Compare commits

..
47 changed files with 1289 additions and 2146 deletions

No files matched your search

-17
View File
@@ -1,17 +0,0 @@
# On-Demand Context Index
Read these files when the task requires their domain knowledge. They are NOT auto-loaded — read them explicitly.
| When to read | File |
|---|---|
| Product vision, business model, Edition system, repository map | `platform-architecture.md` |
| Feature/UX decisions — who is affected, stakeholder impact matrix | `stakeholders-and-human-centered-design.md` |
| Finding files, repo layout, directory tree | `directory-structure.md` |
| Routing a fix to wwv-data-engine, wwv-seeders, marketplace, etc. | `ecosystem-repositories.md` |
| Debugging broken plugin data, namespace collisions, latency, logging | `troubleshooting-and-debugging.md` |
| SSH into server, Coolify MCP operations | `server-management.md` |
| Next.js App Router, data pipeline, Redis architecture, DB schema | `application-architecture.md` |
| `.env` variables and secrets | `environment-config.md` |
| Coding principles, Definition of Done checklist, antipatterns | `coding-principles.md` |
| Boy Scout rule, when to update rules/skills/workflows | `continuous-improvement.md` |
| How to orient and maintain context between sessions | `context-and-memory.md` |
-280
View File
@@ -1,280 +0,0 @@
# Coding Principles
You are about to write or modify code. **Read this file first.** It exists to prevent the failure mode where every fix makes the next harder — the kind of accumulated coupling and band-aid-on-band-aid that turns a clean codebase into a fragile one over months. This file is the constitution that keeps that from happening.
## The hard rule
> **Always interface-based, extensible, composable, modular. Never band-aids on band-aids.**
If you find yourself adding a flag, an `if`-check, or a special case to make a problem go away, you are probably band-aiding. Stop and find the seam where the abstraction broke. Fix it there.
## Core principles
### 1. Interfaces before implementations
Define the contract first. The contract is what other components depend on; the implementation is replaceable. Public functions accept the interface, not the concrete type.
```ts
// Yes
function attachSource(source: DataSource) { ... }
// No
function attachPostgresSource(pg: PostgresInstance) { ... }
```
When a pattern repeats across two modules, refactor it into an interface immediately. Don't wait for the third copy to "earn" the abstraction — by then the costs have compounded.
### 2. Loose coupling through explicit contracts
Components communicate through well-defined contracts: typed function signatures, events on a bus, message queues, or capability registries. Pick the form that fits, but pick *one* and make it the only path.
Components do **not** communicate through:
- Direct imports of each other's internal modules
- Shared mutable state outside the contract
- Synchronous request/response when async fits the actual data flow
- "Just this once" backdoors that bypass the contract
If you're tempted to write `import { internalThing } from '../other-module/internal'`, you are doing it wrong. If the symbol you need isn't part of the other module's public contract, either it should be — propose adding it — or you're solving the problem in the wrong place.
### 3. Domain boundaries are sacred
Each domain (module, service, bounded context, plugin) owns its own state and its own surface. Cross-domain access is **explicit, gated, and audited** — never a shortcut.
Forbidden:
- Reading another domain's storage directly
- Calling another domain's private functions
- Hardcoded `if (tenant === 'X')` or `if (mode === 'special')` branches in shared code
Allowed:
- A documented capability or API on the other domain that you call
- A configuration field on the relevant record that drives the strategy
- Cross-domain access gated by explicit policy (default **deny**, opt-in per case), with the access itself logged
The rule of thumb: if removing a domain would require code archaeology in three other domains to figure out what depends on its internals, the boundary has already been violated. Treat that as a bug.
### 4. Don't bend your architecture to fit a dependency
When you bring in a library or framework, use it where it fits cleanly; build natively where it doesn't.
Let the dependency handle (don't reinvent):
- The thing it's specifically good at, in the shape it expects
Your code owns (the dependency can't represent):
- Anything that's specifically your domain
- Anything that requires identity, lifecycle, or invariants the dependency doesn't model
- The integration seams between the dependency and your domain
If you find yourself bending your design to fit a library convenience, stop. Either the library has a hook for what you need, or you build it natively. Don't compromise architecture for the convenience of pulling in someone else's defaults.
### 5. Data lifetimes are deliberate, not incidental
Different data has different lifetimes. Get this wrong and you'll either retain garbage forever or delete things users care about. Common categories:
- **Operational telemetry** (logs, metrics, internal events) — short rolling window. Useful for crash recovery and immediate investigation; not useful enough to keep forever.
- **User content** (messages, documents, records the user created or that represent their lived experience with the system) — **permanent unless the user explicitly deletes it.** Never pruned by a sweep, never deleted by a retention policy authored for telemetry.
- **Resumable session state** — bounded window appropriate to the use case; archived (state flag flipped) rather than deleted.
Name each category in your codebase. Write the retention policy for each one separately. Never propose a sweep that crosses categories — "delete events older than 30 days" is fine if events are telemetry; the same policy applied to user content is data loss masquerading as housekeeping. If storage pressure ever becomes real for content, introduce cold-tiered storage, not deletion.
### 6. Optional features must be truly removable
Every optional feature (plugin, extension, integration, side module) declares its surface explicitly and communicates through the same contracts as everything else. No optional feature reaches into another's internals.
Test for whether something is core vs. optional: **if removing this feature should break the system, it's core. Otherwise it's optional, and it should be removable without leaving residue.**
If you can't delete an "optional" module's directory and have the system keep running, it isn't optional. Either fix the coupling or move it into core honestly.
### 7. Put code where it belongs
When a feature spans layers (a trigger lives here, the action lives there, the data model lives somewhere else), resist the urge to fold it all into one place "for convenience." The feature lives in three places because each piece belongs to a different domain. Routing across them is the job — not collapsing them.
The general rule: if a feature needs to read/write durable state, it's a tool/extension on top of the layer that owns that state; if it's "the system should know about this concept," it's a configuration or contract change. Often it's both, with a thin handler that wires them together. That thin handler is fine — it's load-bearing — but it doesn't justify smashing the layers together.
## Antipatterns to refuse
These are common failure modes that, repeated over time, calcify a codebase. Refuse them when you catch yourself reaching for them.
| Antipattern | Better answer |
|---|---|
| Adding a new flag to an existing function instead of extracting a strategy | Extract the strategy interface; existing call sites pass the default strategy |
| Patching around a bug at the call site | Find the seam where the abstraction broke; fix it there |
| Direct database access from a feature module | Go through the layer that owns the data; if no surface exists, propose one |
| Hardcoded paths, model names, IDs, environment specifics in business logic | Configuration; or look up via registry |
| `if (tenant === 'X') { ... }` in shared code | Strategy on the record, or a per-tenant config field |
| Inheriting one layer's concerns into another (or vice versa) | Cross the boundary through the documented contract |
| "It works for now" — inline coupling between layers | If the layers are real, draw the contract; if they're not, merge them honestly |
| Adding a defensive `try/catch` to swallow an error | Find why the error happens; gate cleanly at the right boundary |
| Writing a comment to explain what code does | Rename and refactor so the code reads itself |
| Copying a snippet from one module into another "to avoid the dependency" | Either it belongs in a shared module, or the duplicate will drift — pick |
## Debugging discipline
When something breaks:
1. **Reproduce it.** Don't trust intuition about what's failing.
2. **Find the seam.** Where did the abstraction break? Which contract is being violated, and by whom?
3. **Fix at the seam.** Not at the call site. Not at the consumer. At the boundary that's wrong.
4. **If you can't find the seam, surface it.** Tell the user explicitly: "this is a design gap; the right fix needs an architectural decision, not a patch." Do not paper over.
If a fix doesn't work, **don't add another guard on top.** Find the real seam and gate cleanly. Stacked defensive layers are how clean codebases become unmaintainable.
## Lifecycle invariants and state-changing operations
Most recurring bugs cluster around the same shape: an operation crosses lifecycle state without checking, and the result is silent corruption, lost data, or orphan rows. The recipe that prevents recurrences:
1. **Name the invariant.** "Closed sessions don't accept work." "Scoped grants need a scope id." "Subscriptions past retention can't deliver swept events." Write it down before reaching for code.
2. **Throw a typed error at the seam.** The single function that crosses the invariant refuses the bad case with a typed, named error — not a boolean return, not a silent skip, not a generic `Error`.
3. **Pre-check at every call site that mutates *before* reaching the seam.** If the seam check fires after `store.append()`, you've left an orphan row. The pre-check is hygiene; the seam check is correctness — both are required. When the same pre-check appears at a second call site, extract a helper.
4. **Validate, then mutate — never the reverse.** Within a single function, resolve all inputs that could fail (key derivation, lookups, permission checks) before touching durable state. A validation throw should leave nothing half-applied. If you can't separate validation from mutation, the function is doing two jobs.
5. **A documented unsafe API is still unsafe.** A docstring that says "don't use this when X" is a foot-gun, not a warranty disclaimer. Either make the API safe in case X (drain, throw, refuse) or remove it.
If you're tempted to skip step 3 because "the seam check covers it," check whether the call site writes anything before reaching the seam. If yes, you need the pre-check.
## Definition of done — checklist for adding or updating a function
Every code change must satisfy the relevant checklist before merge. Refusal triggers below mean **stop and surface to the user** — do not ship a half-met checklist.
### Adding a new function
All must be true:
- [ ] Belongs in this module (not somewhere else)
- [ ] Fulfills an existing interface, OR a new interface is added with it (and reasons documented in a spec doc / ADR)
- [ ] Has a doc-comment header (purpose + params + returns + thrown errors if relevant)
- [ ] Has at least one test that targets the contract (not implementation details)
- [ ] Test passes locally and in CI
- [ ] Doesn't introduce direct cross-module imports of internals — communication through public contracts only
- [ ] Doesn't touch storage owned by another component — route through that component's surface
- [ ] Matches the spec doc's contract for this component (if no spec covers it, update the spec first)
- [ ] Name is clear enough that the header comment is a summary, not an explanation
### Updating an existing function
All must be true:
- [ ] Reason for change is clear (bug fix / new requirement / refactor / spec change)
- [ ] Doc-comment still accurate after change (update if not)
- [ ] Existing tests cover the new behavior (add new tests if not)
- [ ] Build is green (all tests passing)
- [ ] If this changes the contract: spec doc updated AND an ADR added if the change is load-bearing
- [ ] Fix is at the seam, not at the symptom (no band-aid patches)
- [ ] No new flags or special cases added in lieu of strategy extraction
### Refusal triggers (any one means stop)
- Adding `if (tenant === 'X')` to shared code → use per-tenant config instead
- Adding a flag to make a problem go away → extract the abstraction the flag implies
- Direct DB or filesystem access from a feature module → route through the owning component
- `try`/`catch` swallowing an error → find why the error happens, gate cleanly
- Test omitted because "it's trivial" → no
- Function-header skipped because "it's obvious" → no
- "Just add another guard here" when a guard didn't work → find the real seam
- Patch shipped to "unblock the build" without root-cause fix → no, surface the design gap
If any criterion isn't met, stop. Surface to the user as a design question. Don't paper over.
## Code-style minimums
### Comments
**Every function gets a doc-comment header** in your language's idiomatic format (TSDoc, JSDoc, Python docstrings, Rust `///`, Javadoc, etc.). Not optional. Long-running codebases get re-read after gaps; function-header docs are how re-onboarding stays cheap.
Required content:
- One-line summary at a contract level (what the function guarantees, not what the code does line-by-line)
- Each parameter and what it represents
- Return value
- Thrown errors / failure conditions if relevant
- Example usage when the call shape is non-obvious
Apply this to public, internal, exported, and private functions. Apply to methods on classes too.
**Inline comments inside function bodies** still follow the standard rule: only for non-obvious WHY (hidden constraints, workarounds, surprising behavior, references to bug reports). Don't comment WHAT the code does — well-named identifiers and the function-header doc cover that.
### Tests
**Every function ships with at least one test. Tests passing is a build requirement.**
- Tests target the contract (function signature + documented behavior + invariants), not implementation details. Refactoring should not break tests; behavior changes should.
- A new implementation of an interface must pass the same contract test suite (write tests against the interface, not against a specific implementation).
- Test names describe the case being verified, not the function: `it('refuses resume after archival window expires')` not `it('resumeSession')`.
- For storage code: tests use a real (temporary) instance of the storage layer, created and torn down per test or per suite. Don't mock the storage engine — you're testing the wrong thing if you do.
- For integration code that depends on an external library's hooks: contract tests against that library's surface so a dependency bump catches behavior changes.
- Build pipeline (CI or local) runs all tests; non-zero exit = build fails. There is no "build succeeded but tests failed" state.
### Logging
**Every new feature with a non-happy-path ships with a log line at each rare branch.** Use a structured logger with module-scoped child loggers. The log file is where "something's not working" turns into "what actually happened."
Level guidance:
- **info** — expected-but-noteworthy edges: a UI race that hits a closed resource, a permission bypass for a privileged principal. Correct behavior an operator might want to confirm fired.
- **warn** — should be rare and worth investigating: race past a pre-check, lost events past retention, programmer/operator errors (calling a method after stop, a key that should have resolved didn't).
- **error** — actual failures the operator must respond to: handler crashes, dispatch loop crashes, integration failures.
- **debug** — high-volume internal state changes; always pair with a `module` field so it can be filtered.
Pass structured fields, not interpolated strings — fields are what makes a log queryable:
```ts
// Yes
logger.warn({ session_id, state }, "rejected append against non-live session");
// No
logger.warn(`session ${session_id} (${state}) rejected append`);
```
If you add a new typed error or a new event topic, also log it. The typed error tells the immediate caller; the event tells subscribers; the log tells the operator. The three are independent surfaces for the same fact, and all three are needed — events don't show up in the log, logs don't show up on the bus, errors don't survive the catch.
### Structure
- Keep modules small. If a file is over ~500 lines, ask whether it has more than one responsibility.
- Prefer explicit composition (caller wires things together) over implicit dependency-injection magic.
- One public concept per module. If a file exports both `Foo` and `unrelatedHelperUsedNowhereElse`, the helper has the wrong home.
### Dependencies
**Always check the registry for the current version before adding or updating a dependency.** Don't pick versions from memory.
The simple test: did you actually run `<package-manager> view <pkg> version` (or its equivalent) before pinning? If not, you're guessing — and the guess is almost always behind, sometimes by a major version with active deprecation.
Why this matters:
- **Long-running codebases.** Stale dependency anchors compound. A v3 you picked because that's what was on your mind will still be v3 a year later, and the v4 migration gets harder every month it waits.
- **Active deprecations are real.** Common libraries are mid-deprecation cycles for older majors. Pinning to a soon-deprecated version on day one means the eventual upgrade is forced rather than chosen.
- **"Latest" doesn't mean "use without thinking."** Read the changelog for breaking changes; decide whether to take them now or pin behind explicitly. But the *default* posture should be: pick latest, and justify in writing if pinning behind.
If you must pin behind latest, leave a comment in the manifest (or a CHANGELOG note) explaining why — a known incompatibility with another dep, an in-progress migration, a tested-and-rejected major bump. *"I haven't checked"* is not an acceptable reason.
Apply to: adding a new dep, updating an existing one, scaffolding a new package. *Not* required for routine no-op installs that don't touch versions.
### Doc-vs-spec discipline
Specification documents (the deeper ones that describe how a component must behave) are **functional specifications**, not narrative essays. The discipline:
- **Concrete data structures** — type definitions, schemas, DDL, JSON examples
- **Concrete function signatures** — input/output types, error types
- **Concrete event/message payloads** — name + schema + example
- **Concrete API signatures** — input/output schemas, side effects, permission requirements
- **Flow pseudocode** — when orchestration is non-obvious
- **Invariants** — explicitly listed: "after X happens, Y is always true"
- **Failure modes** — enumerated: "if X fails, the system does Y"
- **Acceptance tests** — statements another engineer can implement: "given input X, output Y"
Avoid in spec docs:
- Long prose explaining motivation (that goes in ADRs / decision records)
- Discussion of alternatives considered (also ADRs)
- Background context (README handles this)
- Marketing/aspirational tone
Spec docs are for implementation. Read one and you should know what to build.
## When in doubt
Ask the human user. Specifically:
- "I don't see a clean seam for this; the options I see are A or B. Which fits the architecture you have in mind?"
- "This requires a new contract / capability / interface. Should I draft it before implementing?"
- "I'm about to add a special case here; that smells like a missing abstraction. Do you want me to extract the abstraction or note the smell and defer?"
Don't ship band-aids and ask forgiveness later. The user has explicitly said: prevent that pattern.
+2 -21
View File
@@ -1,9 +1,7 @@
---
trigger: glob
description: Strict guidelines for rendering geospatial entities, handling 3D terrain clipping, and applying dynamic animation properties against Cesium's runtime engine.
paths:
- "src/core/globe/**/*"
- "src/plugins/**/*"
- "packages/wwv-plugin-*/src/**/*"
globs: "src/core/globe/**/*"
---
# Cesium Rendering Rules
@@ -25,23 +23,6 @@ When creating primitive objects:
- **Points** (`PointPrimitive`): Can use `size`, `outlineWidth`, and `outlineColor`.
- **Billboards** (`BillboardGraphics` for SVG/Icons): Must **NOT** use `size`, `outlineWidth`, or `outlineColor`. You must use `iconScale` to resize them. Mixing point properties on billboards causes severe shader clipping.
## Sizing Conventions
| Entity type | Knob | Default | Typical band | Source |
|-------------|--------------|------------------------------------|------------------|--------|
| `point` | `size` | `8` desktop / `12` mobile | 5–12 | `src/core/globe/primitiveOps.ts` (`defaultPointSize`) |
| `billboard` | `iconScale` | `0.7` (`DEFAULT_BILLBOARD_SCALE`) | 0.6–0.9 | `src/core/globe/primitiveOps.ts` (`DEFAULT_BILLBOARD_SCALE`) |
| `model` | `modelScale` | `1.0` + `modelMinPixelSize` | varies per asset | `src/core/globe/ModelManager.ts`, `src/core/globe/hooks/useModelRendering.ts` |
> [!IMPORTANT]
> **Cross-plugin parity rule.** If your plugin renders alongside maritime / aviation / military-aviation on the globe, **leave `iconScale` unset** (or pick a value inside 0.6–0.9). A value below 0.5 will look visibly broken next to peer plugins. A value above 1.0 will dominate the globe.
> [!CAUTION]
> **Silent-ignore footgun.** `size` is a `PointPrimitive` property. If you set `size` on a `billboard` entity it is **silently dropped** — Cesium does not error and the billboard simply renders at `DEFAULT_BILLBOARD_SCALE`. When converting a point plugin to a billboard plugin, replace `size: N` with `iconScale: ~N/16` (the canvas base is `getBaseSize() = 48px` desktop) — do not delete it and do not translate it literally as `iconScale: N/10`. This footgun has shipped at least once (see commit history on `local-plugins/wwv-plugin-satellite`).
> [!NOTE]
> **Backdrop stack.** `createSvgIconUrl` (in the SDK) already wraps the icon in a 44×44 SVG with a dark backdrop circle. `src/core/globe/iconUpscaler.ts` then redraws onto a 48px canvas with **another** backdrop circle (`DEFAULT_BG = "rgba(15, 23, 42, 0.55)"`). Pass `background: false` to `createSvgIconUrl` when you want only the host's circle. Don't add a third backdrop in your own SVG.
## Performance Culling
> [!TIP]
+1 -6
View File
@@ -1,11 +1,6 @@
---
trigger: model_decision
description: Cloud edition implementation details, PostgreSQL Row-Level Security (RLS) isolation, cross-subdomain auth, and cryptographic license keys.
paths:
- "src/lib/auth*"
- "src/app/api/auth/**/*"
- "src/core/auth.ts"
- "src/core/storage.ts"
- "src/core/tenant.ts"
---
# WorldWideView — Cloud & Auth Architecture
@@ -1,5 +1,5 @@
---
trigger: model_decision
trigger: always_on
---
# Context and Memory Synchronization
@@ -1,5 +1,5 @@
---
trigger: model_decision
trigger: always_on
description: Guidelines for ongoing project maintenance, cleanliness, and self-correction during everyday development tasks.
---
+2 -62
View File
@@ -1,9 +1,6 @@
---
description: Architecture rules and dependency management guidelines for wwv-data-engine seeders, pnpm workspace dependency loading, and WebSocket payload contracts.
paths:
- "packages/**/*"
- "local-seeders/**/*"
- "docker-compose.yml"
trigger: model_decision
description: Architecture rules and dependency management guidelines for wwv-data-engine seeders.
---
# Data Engine V2 Architecture & Dependency Loading
@@ -32,60 +29,3 @@ In production, the `wwv-data-engine-v2` does NOT mount the local file system. In
## 5. Namespace Separation Rule
To prevent collisions and `404` or `ERR_MODULE_NOT_FOUND` errors, seeders **MUST NOT** exist simultaneously in both the community (`wwv-seeders-community`) and private (`wwv-seeders-private`) repositories. Namespace overlaps will cause module resolution failures when the V2 engine attempts to load them. Private seeders have priority.
## 6. Plugin ID Contract — Single Source of Truth
> **See ADR-0002** (`docs/architecture/decisions/adr-0002-seeder-exported-name-as-canonical-plugin-id.md`) for full rationale and alternatives considered.
**Rule:** Every seeder MUST export a `name` field that exactly matches the corresponding frontend plugin's `id` field (kebab-case).
```typescript
// ✅ Correct — seeder src/index.ts
export default {
name: "gps-jamming", // MUST equal the frontend plugin id
cron: "0 0 * * *",
fn: seedGpsJam,
};
// ❌ Wrong — do not use folder names, camelCase, or abbreviations
export default {
name: "gpsjam", // mismatch — frontend plugin is "gps-jamming"
...
};
```
**How the engine uses it:** `seeder-loader.ts` reads each seeder's exported `name` as the canonical plugin ID. This ID is used for:
- The `/manifest` endpoint's `plugins` array (what the frontend checks for local routing)
- WebSocket data messages (`pluginId` field)
- Redis keys (`data:<name>:live`)
- Scheduler logs
**Folder names are organizational only.** The engine warns at startup if `toKebabCase(folderName) !== seeder.name`, but this is cosmetic — the seeder's exported `name` always wins. Folder renames are a future cleanup, not a functional requirement.
**Frontend side:** `resolveEngineUrl(pluginId)` checks `localEngineHasPlugin(pluginId)` against the manifest. If the names match, the plugin routes to `ws://localhost:5000/stream`. No alias maps or translation layers are needed.
## 7. WebSocket Payload Format Contract
When a seeder calls `setLiveSnapshot(pluginId, payload, ttl)`, the engine broadcasts:
```json
{ "type": "data", "pluginId": "my-plugin", "payload": <whatever you passed> }
```
**`WsClient` payload handling rules (in order):**
1. Plugin has `mapWebsocketPayload(payload, existingEntities)` → called with raw payload → must return `GeoEntity[]`
2. No handler + payload **is** a flat `GeoEntity[]` array → used directly (timestamp normalized)
3. No handler + payload is **any other object** → **silently dropped** with a console warning
This means:
| Seeder sends | Frontend needs | Notes |
|---|---|---|
| `[{id, latitude, longitude, ...}]` | Nothing (WsClient handles) | ✅ Simplest for basic point data |
| `{ items: [{...}] }` | `mapWebsocketPayload` | Standard engine snapshot format |
| `{ satellites: [{...}] }` | `mapWebsocketPayload` | Named-collection format |
**Recommendation:** If the seeder and frontend plugin are developed together, use a **flat `GeoEntity[]` array** — no `mapWebsocketPayload` needed. Use named objects (e.g. `{ satellites: [...] }`) only when the seeder data has a domain-specific shape worth preserving; in that case you **must** implement `mapWebsocketPayload` on the frontend plugin.
> [!CAUTION]
> The most common silent bug: seeder sends `{ items: [...] }` (an object), frontend plugin omits `mapWebsocketPayload`. Data arrives but WsClient drops it with a console warning. The globe stays empty. Always implement `mapWebsocketPayload` for any plugin whose seeder sends an object payload.
-6
View File
@@ -1,9 +1,3 @@
---
description: Rules for safe Prisma schema changes, defensive migrations, and PostgreSQL DDL operations.
paths:
- "prisma/**/*"
---
# Database Migrations & Schema Modification Rules
## 1. Defensive Migrations
-9
View File
@@ -1,12 +1,3 @@
---
description: Docker build patterns, pnpm dev commands, Vitest strategy, Coolify deployment rules, and security headers.
paths:
- "Dockerfile"
- "docker-compose.yml"
- ".github/**/*"
- "next.config.ts"
---
# Deployment, Testing, and Operations
## Development Commands
File renamed without changes.
+2 -1
View File
@@ -1,6 +1,7 @@
---
trigger: testing
description: Guidelines and architecture for the Playwright End-to-End (E2E) testing infrastructure.
paths:
globs:
- "tests/**"
- "public/e2e-fixtures/**"
- "playwright.config.ts"
File renamed without changes.
+2 -16
View File
@@ -1,8 +1,6 @@
---
description: Guidelines for the WorldWideView plugin marketplace, database interactions, PKCE auth flow, and dynamic plugin installation.
paths:
- "src/lib/marketplace/**/*"
- "src/app/api/marketplace/**/*"
trigger: model_decision
description: Guidelines for the WorldwideView plugin marketplace, database interactions, and dynamic plugin installation.
---
# Marketplace Architecture
@@ -18,18 +16,6 @@ Plugins from the marketplace use the `bundle` format and are imported at runtime
- `pluginManager.loadFromManifest()` validates and registers these plugins.
- Ensure any schema changes to the marketplace tables follow the `database-migrations.md` rule.
## 4. Fresh-Install Seeding
Fresh installs (no rows in `InstalledPlugin`) auto-seed a starter set so the globe isn't empty on day one. The source of truth is the **signed verified registry** (`marketplace.worldwideview.dev/api/registry`, loaded by `registryClient.getVerifiedPluginIds()`).
- **No hard-coded plugin lists.** Publishing a plugin to the verified registry is what makes it part of the fresh-install starter set.
- **Idempotent**: a `defaults_seeded` row in the `Setting` table prevents re-runs. If the registry is unreachable on first attempt, the guard is NOT set — the next request retries.
- **Existing instances** with `defaults_seeded=true` are untouched. To pull in newly-verified plugins, users install via the marketplace UI.
- **Demo edition** uses an independent path (`NEXT_PUBLIC_DEMO_DEFAULT_PLUGINS` env var, handled in `AppShell.tsx` and `useMarketplaceSync.ts`) — not this seeder.
> [!CAUTION]
> Do not reintroduce a static `DEFAULT_PLUGIN_IDS` constant or any other hard-coded plugin list. That drifted from npm reality, caused 404s for unpublished IDs, and conflated "verified" with "auto-installed."
## Current Implementation State
+2 -4
View File
@@ -1,9 +1,7 @@
---
trigger: model_decision
description: Strict rules for operating within the pnpm monorepo workspace environment structure across frontend and backend services.
paths:
- "packages/**/*"
- "pnpm-workspace.yaml"
- "local-plugins/**/*"
globs: ""
---
# Monorepo Workflow Guidelines
+1 -85
View File
@@ -1,10 +1,6 @@
---
trigger: model_decision
description: The standard operating procedure for instantiating, modifying, and registering a new data source plugin within the engine.
paths:
- "src/core/plugins/**/*"
- "packages/wwv-plugin-*/src/**/*"
- "local-plugins/**/*"
- "packages/wwv-plugin-sdk/src/**/*"
---
# Plugin Architecture & Data Flow
@@ -21,59 +17,6 @@ A valid plugin class MUST provide:
3. `renderEntity(GeoEntity)` to determine visual style.
4. `getPollingInterval()` (defaults fallback securely to standard store config).
## Plugin Data Delivery Types
Plugins receive entity data in one of three modes. Choose the right mode before writing code — mixing them causes silent data loss.
| Mode | `getPollingInterval()` | `fetch()` | `mapWebsocketPayload()` | When to use |
|---|---|---|---|---|
| **REST-only** | `> 0` (ms) | Returns `GeoEntity[]` | Not needed | Static or slow-changing data (< every 30s). No seeder required. |
| **WebSocket-only** | `0` | Return `[]` immediately | **Required** if seeder sends object payload | Live-streaming data from a seeder (satellites, AIS, flights) |
| **Hybrid** | `> 0` | Returns initial snapshot | **Required** if seeder sends object payload | Needs both immediate data on load AND live updates |
### WebSocket-Only Pattern (most live-data plugins)
```typescript
getPollingInterval(): number {
return 0; // WS-only — do NOT call REST on enable
}
async fetch(_timeRange: TimeRange): Promise<GeoEntity[]> {
return []; // Initial data arrives via mapWebsocketPayload within seconds
}
mapWebsocketPayload(payload: any, _existingEntities: GeoEntity[]): GeoEntity[] {
// Transform raw engine payload → GeoEntity[]
// payload shape depends on what the seeder calls setLiveSnapshot() with
const items = payload?.items ?? (Array.isArray(payload) ? payload : []);
return items.map((item: any): GeoEntity => ({
id: `my-plugin-${item.id}`,
pluginId: "my-plugin",
latitude: item.lat,
longitude: item.lon,
altitude: item.alt ?? 0,
timestamp: new Date(),
label: item.name,
properties: item,
}));
}
```
> [!WARNING]
> If `getPollingInterval()` returns `0` but the seeder sends an object payload, you **MUST** implement `mapWebsocketPayload`. Without it, `WsClient` drops the data silently. See `data-engine-architecture.md §7` for the exact rule.
### `mapWebsocketPayload` Signature
```typescript
mapWebsocketPayload(payload: any, existingEntities: GeoEntity[]): GeoEntity[]
```
- `payload` — raw value from `setLiveSnapshot()` in the seeder (whatever shape the seeder stores)
- `existingEntities` — current entities in the store for this plugin (use for incremental updates)
- Must return a **complete replacement** `GeoEntity[]` (not a diff)
---
## Plugin Architectures (Manifest Formats)
Plugins are defined via `PluginManifest` and operate on an **All-Bundle** architecture.
@@ -109,33 +52,6 @@ All plugins MUST define their identity and compatibility via a `"worldwideview"`
}
```
### `peerDependencies` Rule for npm-Published Plugins
Plugins published to npm **MUST** declare the following as `peerDependencies`, not `dependencies`:
```json
{
"peerDependencies": {
"@worldwideview/wwv-plugin-sdk": "*",
"react": ">=19",
"lucide-react": "*"
}
}
```
Also peer (never bundle): `react-dom`, `cesium`, `resium`, `zustand`.
Bundling these causes multiple React instances in the host app, breaking hooks (`Invalid hook call` at runtime).
### File Extension Rule
- Files containing **JSX** (any `<Component />` syntax, `return (<div>...)`) → must use `.tsx`
- Files with no JSX (pure TypeScript, even importing React types) → use `.ts`
This is enforced by the TypeScript compiler. A `.ts` file with JSX will fail to build.
---
## The Registration Pipeline
> [!NOTE]
File renamed without changes.
+2 -3
View File
@@ -1,8 +1,7 @@
---
trigger: glob
description: Guidelines on managing the single source of truth across WorldWideView using Zustand, avoiding prop-drilling, and isolating UI rerenders.
paths:
- "src/core/state/**/*"
- "src/components/**/*"
globs: "src/core/state/**/*, src/components/**/*"
---
# Zustand State Management
@@ -26,29 +26,7 @@ When an AI agent or developer is executing CLI commands via tools in this projec
- **Rule**: NEVER use the `&&` operator to chain commands in PowerShell. It will result in parser errors.
- **Resolution**: Use sequential execution, run commands separately, or use PowerShell-safe chaining like `; if ($?) { }`.
## 5. UTF-8 BOM in `package.json` Crashing Node Scripts
If `pnpm dev` (or any workspace script) dies with `SyntaxError: Unexpected token '', "{ ..." is not valid JSON`, the offending file has a UTF-8 BOM (`EF BB BF`) at byte 0.
- **Root Cause**: Windows PowerShell 5.1 `Set-Content -Encoding UTF8` and Notepad's "Save as UTF-8" both emit a BOM. Node's `fs.readFileSync(p, 'utf-8')` does NOT strip it, so `JSON.parse` rejects the leading ``. pnpm/npm strip BOMs silently, which is why this only blows up in custom workspace scripts.
- **Diagnosis** (PowerShell):
```powershell
Get-ChildItem local-plugins -Directory | ForEach-Object {
$p = Join-Path $_.FullName 'package.json'
if (Test-Path $p) {
$b = [System.IO.File]::ReadAllBytes($p) | Select-Object -First 3
if ($b[0] -eq 0xEF -and $b[1] -eq 0xBB -and $b[2] -eq 0xBF) { $_.Name }
}
}
```
- **Fix at the seam (preferred)**: In any script that parses externally-authored JSON, strip the BOM before parsing. See `readJsonFile()` in [`scripts/sync-local-plugins.mjs`](../../scripts/sync-local-plugins.mjs) as the canonical helper:
```js
let text = fs.readFileSync(filePath, "utf-8");
if (text.charCodeAt(0) === 0xFEFF) text = text.slice(1);
return JSON.parse(text);
```
- **Clean existing files** (one-time): rewrite each file without its first 3 bytes via `[System.IO.File]::WriteAllBytes($p, $bytes[3..($bytes.Length - 1)])`.
- **Prevention**: When authoring files via PowerShell, use `Out-File -Encoding utf8NoBOM` (PS 7+) or `[System.IO.File]::WriteAllText($p, $text, [System.Text.UTF8Encoding]::new($false))` (any PS). Never edit `package.json` in Notepad.
## 6. Next.js vs. Undici Fetch Type Mismatches
## 5. Next.js vs. Undici Fetch Type Mismatches
When implementing SSRF protections (e.g., DNS IP Pinning via custom `Dispatcher`), you will likely use `undici.Agent`.
- **Root Cause**: Passing global Next.js `RequestInit` options directly into `undici.fetch` will throw strict TypeScript errors because the `BodyInit` and `RequestCache` signatures differ, and `dns.lookup` returns a generic `number` instead of exactly `4 | 6` for IP families.
- **Resolution**: Strip out Next.js specific keys, explicitly cast `body` as `any`, and explicitly type `resolvedFamily: number` when interfacing between Next.js request streams and the `undici` library.
@@ -38,7 +38,7 @@ Moves a WorldWideView plugin from the old `packages/` monorepo location (or the
> ```
> Example: `node scripts/migrate-plugin.mjs gps-jamming`
>
> The script will automatically copy the frontend plugin, inject peerDependencies, refactor `index.tsx` routing, and fix the backend seeder's build settings and `SEEDERS_DIR` paths. (Fresh-install seeding no longer involves a hard-coded list — once published to npm with a valid `worldwideview` block, the plugin is registered via the signed verified registry.)
> The script will automatically copy the frontend plugin, inject peerDependencies, refactor `index.tsx` routing, fix the backend seeder's build settings and `SEEDERS_DIR` paths, and register the plugin in `defaultPlugins.ts`.
> **You still need to run `pnpm install` and verify the builds manually as described below.**
> [!IMPORTANT]
@@ -29,15 +29,20 @@ A WorldWideView plugin is a self-contained data source that renders geospatial e
## Architecture Decision
```
Does it need live external API data?
├── NO → Static GeoJSON plugin (wwvStaticCompiler in vite.config.ts)
│ No seeder needed. Place data in data/data.json.
└── YES → Does it update more often than every 30s?
├── NO → Cron Seeder (engine, registerSeeder with cron string)
│ getPollingInterval() = 0, mapWebsocketPayload() required
└── YES → Init Seeder (engine, registerSeeder with init function)
getPollingInterval() = 0, mapWebsocketPayload() required
```dot
digraph plugin_type {
rankdir=TB;
node [shape=diamond];
"Does it need live\nexternal API data?" -> "Is polling\nfrequency > 30s?" [label="yes"];
"Does it need live\nexternal API data?" -> "Static GeoJSON\n(wwvStaticCompiler)" [label="no"];
"Is polling\nfrequency > 30s?" -> "Cron Seeder\n(engine)" [label="yes"];
"Is polling\nfrequency > 30s?" -> "Init Seeder\n(engine)" [label="no"];
"Static GeoJSON\n(wwvStaticCompiler)" [shape=box];
"Cron Seeder\n(engine)" [shape=box];
"Init Seeder\n(engine)" [shape=box];
}
```
---
@@ -141,7 +146,7 @@ export default class WildfiresPlugin implements WorldPlugin {
| `getLayerConfig()` | Layer appearance (color, clustering, max entities) |
| `renderEntity(entity)` | Per-entity Cesium rendering options |
**Optional (but `mapWebsocketPayload` is REQUIRED for WS-only plugins with object payloads):**
**Optional:**
| Method | Purpose |
|---|---|
@@ -153,7 +158,7 @@ export default class WildfiresPlugin implements WorldPlugin {
| `getDetailComponent()` | Custom detail panel for selected entity |
| `getSettingsComponent()` | Plugin settings UI |
| `getGlobeComponent()` | React component injected into globe view |
| `mapWebsocketPayload(payload, existing)` | **Required if seeder payload is an object** (not a flat array). Transforms raw engine payload → `GeoEntity[]`. Without it, `WsClient` drops object payloads silently. |
| `mapWebsocketPayload(payload, existing)` | Transform raw WS data into `GeoEntity[]` |
| `requiresConfiguration(settings)` | Return `true` if plugin needs setup first |
### 1.3 Package Metadata
@@ -513,30 +518,14 @@ export default class IssPlugin implements WorldPlugin {
destroy(): void {}
async fetch(_timeRange: TimeRange): Promise<GeoEntity[]> {
return []; // WS-only — data arrives via mapWebsocketPayload
async fetch(timeRange: TimeRange): Promise<GeoEntity[]> {
return []; // WS-only plugin
}
getPollingInterval(): number {
return 0; // WS-only
}
// REQUIRED: seeder sends { source, fetchedAt, items: [...], totalCount } (an object).
// Without this, WsClient drops the payload silently — the globe stays empty.
mapWebsocketPayload(payload: any, _existingEntities: GeoEntity[]): GeoEntity[] {
const items: any[] = payload?.items ?? (Array.isArray(payload) ? payload : []);
return items.map((item: any): GeoEntity => ({
id: `iss-${item.id}`,
pluginId: "iss",
latitude: item.lat,
longitude: item.lon,
altitude: item.alt ?? 0,
timestamp: new Date(),
label: "International Space Station",
properties: { speed: item.speed },
}));
}
getLayerConfig(): LayerConfig {
return {
color: "#ffffff",
@@ -655,9 +644,7 @@ Categories are **lowercase**: `"aviation"` not `"Aviation"`, `"natural-disaster"
- [ ] **Engine:** Seeder file in `wwv-seeders/src/seeders/<name>.ts` (or private repo)
- [ ] **Engine:** `registerSeeder({ name })` matches frontend plugin `id`
- [ ] **Engine:** Calls `setLiveSnapshot()` with correct plugin ID key
- [ ] **Engine:** Seeder payload is either a flat `GeoEntity[]` array OR plugin implements `mapWebsocketPayload` (any other object is silently dropped by WsClient)
- [ ] **Frontend (WS-only):** `getPollingInterval()` returns `0` and `fetch()` returns `[]`
- [ ] **Frontend (WS-only):** `mapWebsocketPayload(payload, existingEntities)` implemented if seeder sends an object
- [ ] **Engine:** Data shape is an object with `items` array (or plugin implements `mapWebsocketPayload`)
- [ ] **Engine:** SQLite table created in `db.ts` if historical data needed
- [ ] **Verification:** `GET localhost:5000/manifest` includes the plugin ID
- [ ] **Verification:** `GET localhost:5000/health` shows seeder last-run timestamp
-1
View File
@@ -1 +0,0 @@
.agents
+1
View File
@@ -0,0 +1 @@
.agents
-6
View File
@@ -4,11 +4,5 @@ install_deps = "pnpm install"
[[pre-start]]
copy_env = """powershell -Command "if (Test-Path '../worldwideview/.env') { Copy-Item '../worldwideview/.env' '.env' -Force }""""
[[pre-start]]
copy_local_plugins = """powershell -Command "if (Test-Path '../worldwideview/local-plugins') { robocopy '../worldwideview/local-plugins' 'local-plugins' /E /XD node_modules dist .next coverage /XF '*.log' /NFL /NDL /NJH /NJS | Out-Null }""""
[[pre-start]]
copy_local_seeders = """powershell -Command "if (Test-Path '../worldwideview/local-seeders') { robocopy '../worldwideview/local-seeders' 'local-seeders' /E /XD .pnpm-store node_modules /NFL /NDL /NJH /NJS | Out-Null }""""
[[pre-remove]]
docker_clean = "docker compose down -v"
-5
View File
@@ -41,11 +41,6 @@ jobs:
pnpm run predev
- name: Run Playwright tests
env:
# Prevent Firefox from loading the AT-SPI/MSAA accessibility bridge on
# Linux CI — without this, Firefox hangs for ~20 min waiting for a bus
# that doesn't exist on ubuntu-latest runners.
NO_AT_BRIDGE: ${{ matrix.browser == 'firefox' && '1' || '0' }}
run: pnpm test:e2e --project=${{ matrix.browser }}
- uses: actions/upload-artifact@v4
+6 -3
View File
@@ -19,8 +19,13 @@ node_modules/
# agents
/.agents/plans
/.agents/context
/.agents/research/
# claude code IDE config
/.claude/
/.cline/
# performance
/performance/
@@ -59,8 +64,7 @@ yarn-error.log*
.pnpm-debug.log*
# env files (can opt-in for committing if needed)
.env
.env.local
.env*
# vercel
.vercel
@@ -92,7 +96,6 @@ packages/wwv-data-engine/
# IDE workspace files
*.code-workspace
*.vscode
# Build outputs and generated assets
/public/cesium/
+422 -81
View File
@@ -2,8 +2,11 @@
## 1. Project Identity
WorldWideView is a **real-time geospatial intelligence engine** visualizing live global data on an interactive 3D globe. Built with **Next.js 16**, **CesiumJS**, **React 19**, and **Zustand**. Design/feature target: `www.worldmonitor.app` ([reference repo](https://github.com/koala73/worldmonitor)).
WorldWideView is a **real-time geospatial intelligence engine** that visualizes live global data on an interactive 3D globe. Built with **Next.js 16**, **CesiumJS**, **React 19**, and **Zustand**, it renders everything from live aircraft and maritime vessels to conflict events, satellites, and environmental data — all through a modular plugin architecture.
### Target Inspiration
Our primary design, feature-set, and operational layout goal is to mimic the structure and capabilities of `www.worldmonitor.app`.
- **Reference Codebase**: [GitHub - koala73/worldmonitor](https://github.com/koala73/worldmonitor)
---
## 2. Tech Stack
@@ -13,151 +16,489 @@ WorldWideView is a **real-time geospatial intelligence engine** visualizing live
| Framework | Next.js 16 (App Router, `output: "standalone"`) |
| Language | TypeScript 5, strict mode |
| 3D Engine | CesiumJS + Resium (Google Photorealistic 3D Tiles) |
| State | Zustand (9 slices: globe, layers, timeline, ui, filter, data, config, favorites, geojson) |
| State | Zustand (slice-based: globe, layers, timeline, UI, filters, data, config, favorites, geojson) |
| Event Bus | Custom typed `DataBus` (pub/sub singleton) |
| Styling | Vanilla CSS — **no Tailwind** |
| Database | PostgreSQL via Prisma 7 |
| Database | PostgreSQL via Prisma 7 (local and cloud) |
| Auth | NextAuth v5 beta (Credentials provider, JWT sessions) |
| Package Manager | pnpm (monorepo with `pnpm-workspace.yaml`) |
| Testing | Vitest + jsdom + React Testing Library |
| Deployment | Docker multi-stage build → Coolify |
| Analytics | Vercel Analytics / custom `trackEvent` |
---
## 3. Critical Invariants
## 3. Directory Structure
Agents MUST respect these at all times:
> [!NOTE]
> See `.agents/rules/directory-structure.md` for the full project directory map and related repositories.
---
## 4. Architecture Patterns
> [!NOTE]
> See [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) for the human-facing architecture deep-dive — plugin system, state management, data pipeline, data engine & seeders, rendering pipeline, and edition system.
Key invariants agents MUST respect:
- **Plugin source of truth**: `@worldwideview/wwv-plugin-sdk` — never define plugin types locally.
- **All-Bundle Model**: every plugin is dynamically imported via `loadPluginFromManifest` using `import(/* webpackIgnore: true */ entry)`. Legacy `StaticDataPlugin` / `DeclarativePlugin` runtimes are deprecated.
- **Agnostic frontend**: each plugin **MUST declare its own `streamUrl`**; do not assume one shared pipe.
- **Nine Zustand slices** under `src/core/state/`: access via `useStore` in React, `useStore.getState()` elsewhere.
- **Primitive-based rendering**: Point/Billboard/Label/Polyline collections only. Never mix `size`/`outlineWidth`/`outlineColor` onto billboard entities — GPU silently clips.
- **Three editions** via `NEXT_PUBLIC_WWV_EDITION` (`local` / `cloud` / `demo`); feature flags in `src/core/edition.ts`.
- **All-Bundle Model**: every plugin (seeder, CDN-loaded, static-compiled, active-proxied) is dynamically imported via `loadPluginFromManifest` using `import(/* webpackIgnore: true */ entry)`. Legacy `StaticDataPlugin` / `DeclarativePlugin` runtimes are deprecated.
- **Agnostic frontend**: the renderer has no concept of a "unified" data engine. Each plugin **MUST declare its own `streamUrl`** in its manifest or config; do not assume one shared pipe.
- **Nine Zustand slices** under `src/core/state/`: `globe`, `layers`, `timeline`, `ui`, `filter`, `data`, `config`, `favorites`, `geojson`. Access via `useStore` in React, `useStore.getState()` elsewhere.
- **Primitive-based rendering** (not Cesium Entity API). Point/Billboard/Label/Polyline collections only. Never mix `size`/`outlineWidth`/`outlineColor` onto billboard entities — GPU silently clips.
- **Three editions** controlled by `NEXT_PUBLIC_WWV_EDITION` (`local` / `cloud` / `demo`); feature flags derived in `src/core/edition.ts`.
---
## 4. Conventions
## 5. Critical Conventions
- **File size**: ~300 lines max. Extract helpers, split components, use hooks.
- **Import aliases**: `@/*` → `./src/*`; `@worldwideview/wwv-plugin-sdk` → `./packages/wwv-plugin-sdk/src`
- **CSS**: Vanilla CSS only. Global: `src/app/globals.css`. Scoped: CSS Modules. HUD: `src/styles/hud-animations.css`.
- **Rendering entities**: Points use `type: "point"` + `size`/`outlineColor`/`outlineWidth`. Billboards use `type: "billboard"` + `iconUrl`/`iconScale`. NEVER mix.
- **Plugin registration**: Built-ins via `AppShell.tsx` → `PluginRegistry` → `PluginManager`. Marketplace plugins via `InstalledPluginsLoader`.
- **Workspace**: Use `"workspace:*"` (not `"*"`) for internal deps. New `packages/` plugins need `transpilePackages` in `next.config.ts`.
- **Temp files**: Save debugging scripts/outputs exclusively in `/local-scripts/` — never in root.
- **Cleanliness**: Remove dead code, unused imports, debug `console.log` before finalising. Never use `any` or `@ts-ignore`. Never create `.mdc` files.
### 5.1 File Size
---
**Guideline: ~300 lines per file.** If a file grows excessively large, consider modularizing it. Extract helpers, split components, or use hooks.
## 5. AI Meta-Directives
### 5.2 Import Aliases
- `@/*` → `./src/*`
- `@worldwideview/wwv-plugin-sdk` → `./packages/wwv-plugin-sdk/src`
- Plugin modules are dynamically imported and do not require individual `tsconfig.json` aliases
### 5.3 CSS Rules
- **Vanilla CSS only** — no Tailwind, no CSS-in-JS
- Global styles in `src/app/globals.css`
- Component-scoped styles use CSS Modules (`.module.css`) or co-located `.css` files
- HUD animations in `src/styles/hud-animations.css`
### 5.4 Rendering Entity Rules
When returning `CesiumEntityOptions` from `renderEntity()`:
- **Points**: Use `type: "point"` with `color`, `size`, `outlineColor`, `outlineWidth`
- **Billboards**: Use `type: "billboard"` with `iconUrl`, `color`, `iconScale`
- **NEVER mix**: Do not use `size`/`outlineWidth`/`outlineColor` on billboard entities — causes GPU clipping
### 5.5 Plugin Registration
Built-in plugins are instantiated in `AppShell.tsx` and registered via `PluginRegistry` → `PluginManager`. Marketplace-installed plugins are loaded from the database via `InstalledPluginsLoader`.
### 5.6 Workspace Rules
- Always run `pnpm install` from project root after creating new packages or linking local plugins
- Official plugin packages go in `packages/wwv-plugin-<name>/`
- Experimental/local plugin sandboxes go in `local-plugins/wwv-plugin-<name>/`
- Globs for `packages/*`, `packages/*/backend`, and `local-plugins/*` are mapped in `pnpm-workspace.yaml` and `tsconfig.json` paths
- Add new `packages/` plugins to `transpilePackages` in `next.config.ts` if required
### 5.7 AI Meta-Directives: Antigravity Standard
> [!NOTE]
> This repository is orchestrated via the **Antigravity open standard**. The entry point for the agent framework is `CLAUDE.md` / `AGENTS.md` at the project root.
> [!WARNING]
> - Always use `.md` extensions. Never `.mdc`. Never reference Cursor IDE rules.
> - **MUST** bump semver in `package.json` before every commit via `/commit` workflow (`feat:` → Minor, `fix/refactor/perf:` → Patch).
> - **MUST** explain complex concepts simply — include an everyday-life analogy.
> - **MUST** require explicit user authorization before any state-changing action that isn't simple/safe.
> - **MUST** ask clarifying questions rather than assume when requirements are unclear.
> - **MUST** update `.agents/rules/` files immediately whenever an architectural shift invalidates them.
> - **Always** use standard `.md` file extensions for rules, skills, and workflows.
> - **Never** use proprietary `.mdc` extensions.
> - **Never** reference Cursor IDE rules; we use the open `.agents/` standard.
> - **MUST**: You MUST update Semantic Versioning numbering inside the relevant `package.json` file prior to executing any code commits, adhering strictly to the `[/commit]` workflow rules (`feat:` -> Minor, `fix/refactor/perf:` -> Patch).
> - **MUST Detail Commit Levels & Bumps**: On description changes or release notes, you must detail the level of commit (Major/Minor/Fix) for *each* individual change. If there are multiple accumulated changes, you MUST EITHER commit them individually and bump the version each time, OR commit them all at once and bump the version multiple times.
> - **MUST Explain Complex Concepts Simply**: Whenever providing a complicated technical explanation to the user, you MUST include a simple explanation below it. Use an analogy with reference to the correct terminology, comparing the concept to something from everyday life to ensure the user easily understands it.
> - **MUST Be Transparent & Narrate Actions (Gemini 3.1 Agents)**: If you are a Gemini 3.1 agent, you MUST always be fully transparent. **Whenever you do anything, you must explicitly say what you are going to do in a visible chat message to the user, and ONLY THEN do it.** This ensures the user can actually see and verify what you are doing in real time. Do not jump to destructive actions without stating your intent first. Keep this narration conversational and natural. Avoid stiff, robotic templates. Just explicitly drop a brief, casual note about what you are checking, fixing, or deleting *before* you run the tool.
> - **MUST Require Explicit Authorization**: Do NOT execute any state-changing tools (e.g., modifying files, writing code, running scripts, executing git commands) without first obtaining clear, explicit permission from the user to proceed with that specific action, **UNLESS** the action is a simple, straightforward, and safe operation (e.g., no deletion or dangerous operations). For safe operations, you may proceed without explicit permission. Otherwise, inform the user of your proposed plan and WAIT for their approval before acting. **Crucially, if you realize you have violated this rule by taking an unauthorized action, DO NOT automatically revert it. Reverting is itself an action that requires authorization. Just answer the question.**
> - **MUST Ask Clarifying Questions**: Never assume anything. If requirements are unclear, if you encounter an unexpected roadblock, or if the user's intent could be interpreted in multiple ways, you MUST pause and ask clarifying questions. Do NOT proceed until the developer has explicitly answered your question or requested you to proceed anyway.
### 5.8 Workspace Hygiene
Whenever agents generate temporary debugging scripts, test REST endpoints via `.mjs`, or dump traces/JSON outputs, they **MUST** save these exclusively inside `/local-scripts/`. The root directory is strictly for production configuration files.
### 5.9 Rule Adherence & Proactive Updates
> [!CAUTION]
> **Agents will trust the `.agents/rules/` files unconditionally.**
> Whenever you execute a major architectural shift or make a change that invalidates an existing rule, you **MUST** immediately update the corresponding rule file(s) and/or `AGENTS.md` to reflect the new reality. Do not leave rules untouched after a change; operating on outdated information will cause future agents to break the codebase.
---
## 6. Environment & Configuration
See `.agents/context/environment-config.md` for required environment variables and secrets.
> [!NOTE]
> See `.agents/rules/environment-config.md` for required environment variables and secrets.
---
## 7. Development & Deployment
## 7. Development, Deployment & Testing
```bash
pnpm dev # Frontend only (auto-runs prisma db push + copy-cesium)
pnpm dev:all # Frontend + data engine via Docker Compose
pnpm build # Production build
pnpm test # Vitest
pnpm db:reset # Wipe + re-migrate DB (destructive)
```
See `.agents/context/` → [deployment and testing details in `.agents/rules/deployment-and-testing.md`] for Docker architecture, Coolify rules, and CSP headers.
> [!NOTE]
> See `.agents/rules/deployment-and-testing.md` for development commands, Docker architecture, Vitest strategy, and CSP security headers.
---
## 8. On-Demand Rules (path-scoped, auto-load on file access)
## 12. On-Demand Rules
These load automatically when you read/edit files matching their paths:
Read the relevant rule file when working in that domain:
| Rule | Triggers on |
|---|---|
| `.agents/rules/cesium-rendering.md` | `src/core/globe/**`, `src/plugins/**`, `packages/wwv-plugin-*/src/**` |
| `.agents/rules/state-management.md` | `src/core/state/**`, `src/components/**` |
| `.agents/rules/plugin-architecture.md` | `src/core/plugins/**`, `packages/wwv-plugin-*/src/**`, `local-plugins/**` |
| `.agents/rules/marketplace-architecture.md` | `src/lib/marketplace/**`, `src/app/api/marketplace/**` |
| `.agents/rules/cloud-auth-architecture.md` | `src/lib/auth*`, `src/app/api/auth/**`, `src/core/auth.ts` |
| `.agents/rules/database-migrations.md` | `prisma/**` |
| `.agents/rules/monorepo-workflow.md` | `packages/**`, `pnpm-workspace.yaml`, `local-plugins/**` |
| `.agents/rules/data-engine-architecture.md` | `packages/**`, `local-seeders/**`, `docker-compose.yml` |
| `.agents/rules/deployment-and-testing.md` | `Dockerfile`, `docker-compose.yml`, `.github/**`, `next.config.ts` |
| `.agents/rules/e2e-testing.md` | `tests/**`, `public/e2e-fixtures/**`, `playwright.config.ts` |
| Rule | When to use | Path |
|---|---|---|
| `platform-architecture` | High-level platform goals, product vision, business model, and Edition System | `.agents/rules/platform-architecture.md` |
| `application-architecture` | Next.js frontend, Zustand state management, and CesiumJS integration | `.agents/rules/application-architecture.md` |
| `plugin-architecture` | Creating/modifying plugins, lifecycle, capability declarations, and seeders (See ADR-0001) | `.agents/rules/plugin-architecture.md` |
| `marketplace-architecture` | Dynamic plugin installation, DB sync, CDN loading, and PKCE Auth (See ADR-0001) | `.agents/rules/marketplace-architecture.md` |
| `cloud-auth-architecture` | Cloud edition, PostgreSQL RLS, multi-tenancy, and licensing | `.agents/rules/cloud-auth-architecture.md` |
| `server-management` | Server development and debugging using SSH and Coolify MCP | `.agents/rules/server-management.md` |
| `stakeholders-and-human-centered-design` | Human-centered design principles and stakeholder map | `.agents/rules/stakeholders-and-human-centered-design.md` |
| `directory-structure` | Project structure and related repositories | `.agents/rules/directory-structure.md` |
| `ecosystem-repositories` | WorldWideView ecosystem repositories and cross-repository workflows | `.agents/rules/ecosystem-repositories.md` |
| `deployment-and-testing` | Docker build patterns, testing strategies, security headers | `.agents/rules/deployment-and-testing.md` |
| `environment-config` | Secrets and environment variables | `.agents/rules/environment-config.md` |
| `monorepo-workflow` | pnpm commands, adding packages, workspace config | `.agents/rules/monorepo-workflow.md` |
| `data-engine-architecture` | Data Engine backend seeder loading, pnpm workspace dependencies | `.agents/rules/data-engine-architecture.md` |
| `cesium-rendering` | Globe rendering, entity types, primitives, LOD, culling | `.agents/rules/cesium-rendering.md` |
| `state-management` | Zustand slices, store access, plugin settings | `.agents/rules/state-management.md` |
| `database-migrations` | Prisma schema changes, migrations, PostgreSQL | `.agents/rules/database-migrations.md` |
| `continuous-improvement` | When to create/update rules, skills, or workflows | `.agents/rules/continuous-improvement.md` |
| `context-and-memory` | How to orient and maintain project context between sessions | `.agents/rules/context-and-memory.md` |
| `troubleshooting-and-debugging` | Resolving plugin latency, namespace collisions, or deployment issues | `.agents/rules/troubleshooting-and-debugging.md` |
| `e2e-testing` | Guidelines and architecture for the Playwright End-to-End (E2E) testing infrastructure | `.agents/rules/e2e-testing.md` |
---
## 9. On-Demand Context (read explicitly when needed)
## 13. Slash Commands / Workflows
See `.agents/context/INDEX.md` for the full navigation table. Key files:
| When to read | File |
|---|---|
| Product vision, business model, Edition system | `.agents/context/platform-architecture.md` |
| Finding files, repo layout | `.agents/context/directory-structure.md` |
| Routing fix to correct repository | `.agents/context/ecosystem-repositories.md` |
| Next.js, data pipeline, Redis, DB schema | `.agents/context/application-architecture.md` |
| Debugging plugin data, namespace collisions | `.agents/context/troubleshooting-and-debugging.md` |
| SSH, Coolify MCP | `.agents/context/server-management.md` |
| Coding principles, Definition of Done | `.agents/context/coding-principles.md` |
| `.env` variables and secrets | `.agents/context/environment-config.md` |
---
## 10. Slash Commands
Invoke by name. Read the skill/workflow file and follow its steps.
| Command | Description | File |
|---|---|---|
| `/commit` | **Required before every commit** — bump semver + conventional commit | `.agents/skills/commit/SKILL.md` |
| `/remember` | Save a lesson or fact into permanent memory | `.agents/skills/remember/SKILL.md` |
| `/remember` | Save a lesson, constraint, or fact into `.agents/` permanent memory | `.agents/skills/remember/SKILL.md` |
| `/pr-review` | 6-role comprehensive pull request review | `.agents/skills/pr-review/SKILL.md` |
| `/local-dev` | Check, start, troubleshoot local dev environment | `.agents/workflows/local-dev.md` |
| `/update-context` | Sync `.agents/context/` with current project state | Global skill |
| `/local-dev` | Check, start, and troubleshoot local dev environment | `.agents/workflows/local-dev.md` |
| `/data-engine-cli` | Use the wwv-data-engine CLI wrapper | `.agents/workflows/data-engine-cli.md` |
| `/debugging-coolify` | Troubleshoot deployed apps on Coolify via MCP/SSH | `.agents/workflows/debugging-coolify.md` |
| `/five` | Five Whys root cause analysis | `.agents/workflows/five.md` |
| `/stitch-to-nextjs` | Generate UI with Stitch MCP, port into Next.js | `.agents/workflows/stitch-to-nextjs.md` |
| `/bing-news-hydration` | Hydrate event attributes with Bing RSS news | `.agents/workflows/bing-news-hydration.md` |
| `/generate-user-roadmap` | Generate updated user-facing roadmap | `.agents/workflows/generate-user-roadmap.md` |
---
## 11. Agent Skills Reference
## 14. Agent Skills Reference
Refer to these skill documents for specialized tasks:
### Project Skills (`.agents/skills/`)
| Skill | When to Use |
|---|---|
| `worldwideview-plugin-creation` | **Use when creating any plugin** — strict architectural checklist |
| `plugin-creation-master-guide.md` | Decision matrix for choosing plugin architecture |
| `osm-static-plugin-creation.md` | Creating static GeoJSON plugins from OpenStreetMap |
| `microservice-plugin-creation.md` | Legacy guide for standalone Fastify microservices |
| `database-operations.md` | Prisma schema changes, migrations, database queries |
| `database-incident-recovery-procedures.md` | Safely restoring a broken production database |
| `database-incident-recovery-procedures.md` | Authoritative protocol for safely restoring a broken production database |
| `index-documentation.md` | Maintaining project documentation index |
| `context7` | Fetch up-to-date library docs via Context7 API |
| `cesium-context7` | CesiumJS-specific documentation lookup |
52 global skills available. See `.agents/global-skills-index.md`.
### Global Skills
52 skills are available across all projects. See `.agents/global-skills-index.md` for the full list and invocation paths.
---
## 12. Pull Request & Commit Guidelines
## 15. Pull Request & Commit Guidelines
- **Commit format**: Conventional Commits (`feat:`, `fix:`, `refactor:`, `perf:`). Use `/commit` workflow.
- **Required checks**: `pnpm test` and `pnpm build` must pass before merge.
- **Review**: Use `/pr-review` for comprehensive multi-role review.
- **Worktrees**: Use `git-wt switch --create <branch>` and `git-wt remove` (never `rm -rf` a worktree — orphans the PostgreSQL Docker volume).
- **Commit Format**: We strictly enforce Conventional Commits (`feat:`, `fix:`, `refactor:`, `perf:`).
- **Workflow**: You **MUST** use the `/commit` workflow before every git commit to ensure proper semantic versioning bumps.
- **Required Checks**: Ensure `pnpm test` and `pnpm build` complete successfully before proposing a merge.
- **Review Process**: Use `/pr-review` to conduct a comprehensive multi-role review on any pull request.
- **Git Worktrees & Database Lifecycle**: We use [Worktrunk](https://github.com/max-sixty/worktrunk) to manage parallel branches and avoid relative path breakage. `git-wt switch --create <branch>` provisions isolated sibling worktrees configured with `.env` files and `pnpm install`. The database port is assigned deterministically based on the directory hash.
> [!CAUTION]
> You MUST use `git-wt remove` to clean up worktrees. Manually deleting a worktree folder (`rm -rf`) will orphan the associated PostgreSQL Docker volume. If this happens, you must manually run `pnpm run db:prune` in the main repository to reclaim disk space.
---
## Hard Rule
## 16. Debugging and Troubleshooting
- **Prisma & PostgreSQL Sync Issues**: If the local database state falls out of sync with Prisma schema, do not manually drop tables. Instead, run `pnpm db:reset` to cleanly wipe and re-apply all migrations.
- **Cesium Entity Clipping**: If `billboard` entities are clipping or failing to render correctly, verify that you are not mixing `point` primitive properties (like `size` or `outlineWidth`) into the `billboard` options.
- **Build Exhaustion (Docker)**: Multi-stage pnpm builds generate massive cache layers. If a Coolify deployment fails silently or PostgreSQL crashes abruptly, check host disk space and run `docker builder prune -a -f`.
- **Next.js Typechecking Failures**: If backend-only `scripts/` fail during Next.js build, verify that the scripts directory is properly listed in the `exclude` array of `tsconfig.json`.
## 17. Coding Principles
You are about to write or modify code. **Read this file first.** It exists to prevent the failure mode where every fix makes the next harder — the kind of accumulated coupling and band-aid-on-band-aid that turns a clean codebase into a fragile one over months. This file is the constitution that keeps that from happening.
### The hard rule
> **Always interface-based, extensible, composable, modular. Never band-aids on band-aids.**
Read `.agents/context/coding-principles.md` before any non-trivial code change.
If you find yourself adding a flag, an `if`-check, or a special case to make a problem go away, you are probably band-aiding. Stop and find the seam where the abstraction broke. Fix it there.
### Core principles
#### 1. Interfaces before implementations
Define the contract first. The contract is what other components depend on; the implementation is replaceable. Public functions accept the interface, not the concrete type.
```ts
// Yes
function attachSource(source: DataSource) { ... }
// No
function attachPostgresSource(pg: PostgresInstance) { ... }
```
When a pattern repeats across two modules, refactor it into an interface immediately. Don't wait for the third copy to "earn" the abstraction — by then the costs have compounded.
#### 2. Loose coupling through explicit contracts
Components communicate through well-defined contracts: typed function signatures, events on a bus, message queues, or capability registries. Pick the form that fits, but pick *one* and make it the only path.
Components do **not** communicate through:
- Direct imports of each other's internal modules
- Shared mutable state outside the contract
- Synchronous request/response when async fits the actual data flow
- "Just this once" backdoors that bypass the contract
If you're tempted to write `import { internalThing } from '../other-module/internal'`, you are doing it wrong. If the symbol you need isn't part of the other module's public contract, either it should be — propose adding it — or you're solving the problem in the wrong place.
#### 3. Domain boundaries are sacred
Each domain (module, service, bounded context, plugin) owns its own state and its own surface. Cross-domain access is **explicit, gated, and audited** — never a shortcut.
Forbidden:
- Reading another domain's storage directly
- Calling another domain's private functions
- Hardcoded `if (tenant === 'X')` or `if (mode === 'special')` branches in shared code
Allowed:
- A documented capability or API on the other domain that you call
- A configuration field on the relevant record that drives the strategy
- Cross-domain access gated by explicit policy (default **deny**, opt-in per case), with the access itself logged
The rule of thumb: if removing a domain would require code archaeology in three other domains to figure out what depends on its internals, the boundary has already been violated. Treat that as a bug.
#### 4. Don't bend your architecture to fit a dependency
When you bring in a library or framework, use it where it fits cleanly; build natively where it doesn't.
Let the dependency handle (don't reinvent):
- The thing it's specifically good at, in the shape it expects
Your code owns (the dependency can't represent):
- Anything that's specifically your domain
- Anything that requires identity, lifecycle, or invariants the dependency doesn't model
- The integration seams between the dependency and your domain
If you find yourself bending your design to fit a library convenience, stop. Either the library has a hook for what you need, or you build it natively. Don't compromise architecture for the convenience of pulling in someone else's defaults.
#### 5. Data lifetimes are deliberate, not incidental
Different data has different lifetimes. Get this wrong and you'll either retain garbage forever or delete things users care about. Common categories:
- **Operational telemetry** (logs, metrics, internal events) — short rolling window. Useful for crash recovery and immediate investigation; not useful enough to keep forever.
- **User content** (messages, documents, records the user created or that represent their lived experience with the system) — **permanent unless the user explicitly deletes it.** Never pruned by a sweep, never deleted by a retention policy authored for telemetry.
- **Resumable session state** — bounded window appropriate to the use case; archived (state flag flipped) rather than deleted.
Name each category in your codebase. Write the retention policy for each one separately. Never propose a sweep that crosses categories — "delete events older than 30 days" is fine if events are telemetry; the same policy applied to user content is data loss masquerading as housekeeping. If storage pressure ever becomes real for content, introduce cold-tiered storage, not deletion.
#### 6. Optional features must be truly removable
Every optional feature (plugin, extension, integration, side module) declares its surface explicitly and communicates through the same contracts as everything else. No optional feature reaches into another's internals.
Test for whether something is core vs. optional: **if removing this feature should break the system, it's core. Otherwise it's optional, and it should be removable without leaving residue.**
If you can't delete an "optional" module's directory and have the system keep running, it isn't optional. Either fix the coupling or move it into core honestly.
#### 7. Put code where it belongs
When a feature spans layers (a trigger lives here, the action lives there, the data model lives somewhere else), resist the urge to fold it all into one place "for convenience." The feature lives in three places because each piece belongs to a different domain. Routing across them is the job — not collapsing them.
The general rule: if a feature needs to read/write durable state, it's a tool/extension on top of the layer that owns that state; if it's "the system should know about this concept," it's a configuration or contract change. Often it's both, with a thin handler that wires them together. That thin handler is fine — it's load-bearing — but it doesn't justify smashing the layers together.
### Antipatterns to refuse
These are common failure modes that, repeated over time, calcify a codebase. Refuse them when you catch yourself reaching for them.
| Antipattern | Better answer |
|---|---|
| Adding a new flag to an existing function instead of extracting a strategy | Extract the strategy interface; existing call sites pass the default strategy |
| Patching around a bug at the call site | Find the seam where the abstraction broke; fix it there |
| Direct database access from a feature module | Go through the layer that owns the data; if no surface exists, propose one |
| Hardcoded paths, model names, IDs, environment specifics in business logic | Configuration; or look up via registry |
| `if (tenant === 'X') { ... }` in shared code | Strategy on the record, or a per-tenant config field |
| Inheriting one layer's concerns into another (or vice versa) | Cross the boundary through the documented contract |
| "It works for now" — inline coupling between layers | If the layers are real, draw the contract; if they're not, merge them honestly |
| Adding a defensive `try/catch` to swallow an error | Find why the error happens; gate cleanly at the right boundary |
| Writing a comment to explain what code does | Rename and refactor so the code reads itself |
| Copying a snippet from one module into another "to avoid the dependency" | Either it belongs in a shared module, or the duplicate will drift — pick |
### Debugging discipline
When something breaks:
1. **Reproduce it.** Don't trust intuition about what's failing.
2. **Find the seam.** Where did the abstraction break? Which contract is being violated, and by whom?
3. **Fix at the seam.** Not at the call site. Not at the consumer. At the boundary that's wrong.
4. **If you can't find the seam, surface it.** Tell the user explicitly: "this is a design gap; the right fix needs an architectural decision, not a patch." Do not paper over.
If a fix doesn't work, **don't add another guard on top.** Find the real seam and gate cleanly. Stacked defensive layers are how clean codebases become unmaintainable.
### Lifecycle invariants and state-changing operations
Most recurring bugs cluster around the same shape: an operation crosses lifecycle state without checking, and the result is silent corruption, lost data, or orphan rows. The recipe that prevents recurrences:
1. **Name the invariant.** "Closed sessions don't accept work." "Scoped grants need a scope id." "Subscriptions past retention can't deliver swept events." Write it down before reaching for code.
2. **Throw a typed error at the seam.** The single function that crosses the invariant refuses the bad case with a typed, named error — not a boolean return, not a silent skip, not a generic `Error`.
3. **Pre-check at every call site that mutates *before* reaching the seam.** If the seam check fires after `store.append()`, you've left an orphan row. The pre-check is hygiene; the seam check is correctness — both are required. When the same pre-check appears at a second call site, extract a helper.
4. **Validate, then mutate — never the reverse.** Within a single function, resolve all inputs that could fail (key derivation, lookups, permission checks) before touching durable state. A validation throw should leave nothing half-applied. If you can't separate validation from mutation, the function is doing two jobs.
5. **A documented unsafe API is still unsafe.** A docstring that says "don't use this when X" is a foot-gun, not a warranty disclaimer. Either make the API safe in case X (drain, throw, refuse) or remove it.
If you're tempted to skip step 3 because "the seam check covers it," check whether the call site writes anything before reaching the seam. If yes, you need the pre-check.
### Definition of done — checklist for adding or updating a function
Every code change must satisfy the relevant checklist before merge. Refusal triggers below mean **stop and surface to the user** — do not ship a half-met checklist.
#### Adding a new function
All must be true:
- [ ] Belongs in this module (not somewhere else)
- [ ] Fulfills an existing interface, OR a new interface is added with it (and reasons documented in a spec doc / ADR)
- [ ] Has a doc-comment header (purpose + params + returns + thrown errors if relevant)
- [ ] Has at least one test that targets the contract (not implementation details)
- [ ] Test passes locally and in CI
- [ ] Doesn't introduce direct cross-module imports of internals — communication through public contracts only
- [ ] Doesn't touch storage owned by another component — route through that component's surface
- [ ] Matches the spec doc's contract for this component (if no spec covers it, update the spec first)
- [ ] Name is clear enough that the header comment is a summary, not an explanation
#### Updating an existing function
All must be true:
- [ ] Reason for change is clear (bug fix / new requirement / refactor / spec change)
- [ ] Doc-comment still accurate after change (update if not)
- [ ] Existing tests cover the new behavior (add new tests if not)
- [ ] Build is green (all tests passing)
- [ ] If this changes the contract: spec doc updated AND an ADR added if the change is load-bearing
- [ ] Fix is at the seam, not at the symptom (no band-aid patches)
- [ ] No new flags or special cases added in lieu of strategy extraction
#### Refusal triggers (any one means stop)
- Adding `if (tenant === 'X')` to shared code → use per-tenant config instead
- Adding a flag to make a problem go away → extract the abstraction the flag implies
- Direct DB or filesystem access from a feature module → route through the owning component
- `try`/`catch` swallowing an error → find why the error happens, gate cleanly
- Test omitted because "it's trivial" → no
- Function-header skipped because "it's obvious" → no
- "Just add another guard here" when a guard didn't work → find the real seam
- Patch shipped to "unblock the build" without root-cause fix → no, surface the design gap
If any criterion isn't met, stop. Surface to the user as a design question. Don't paper over.
### Code-style minimums
#### Comments
**Every function gets a doc-comment header** in your language's idiomatic format (TSDoc, JSDoc, Python docstrings, Rust `///`, Javadoc, etc.). Not optional. Long-running codebases get re-read after gaps; function-header docs are how re-onboarding stays cheap.
Required content:
- One-line summary at a contract level (what the function guarantees, not what the code does line-by-line)
- Each parameter and what it represents
- Return value
- Thrown errors / failure conditions if relevant
- Example usage when the call shape is non-obvious
Apply this to public, internal, exported, and private functions. Apply to methods on classes too.
**Inline comments inside function bodies** still follow the standard rule: only for non-obvious WHY (hidden constraints, workarounds, surprising behavior, references to bug reports). Don't comment WHAT the code does — well-named identifiers and the function-header doc cover that.
#### Tests
**Every function ships with at least one test. Tests passing is a build requirement.**
- Tests target the contract (function signature + documented behavior + invariants), not implementation details. Refactoring should not break tests; behavior changes should.
- A new implementation of an interface must pass the same contract test suite (write tests against the interface, not against a specific implementation).
- Test names describe the case being verified, not the function: `it('refuses resume after archival window expires')` not `it('resumeSession')`.
- For storage code: tests use a real (temporary) instance of the storage layer, created and torn down per test or per suite. Don't mock the storage engine — you're testing the wrong thing if you do.
- For integration code that depends on an external library's hooks: contract tests against that library's surface so a dependency bump catches behavior changes.
- Build pipeline (CI or local) runs all tests; non-zero exit = build fails. There is no "build succeeded but tests failed" state.
#### Logging
**Every new feature with a non-happy-path ships with a log line at each rare branch.** Use a structured logger with module-scoped child loggers. The log file is where "something's not working" turns into "what actually happened."
Level guidance:
- **info** — expected-but-noteworthy edges: a UI race that hits a closed resource, a permission bypass for a privileged principal. Correct behavior an operator might want to confirm fired.
- **warn** — should be rare and worth investigating: race past a pre-check, lost events past retention, programmer/operator errors (calling a method after stop, a key that should have resolved didn't).
- **error** — actual failures the operator must respond to: handler crashes, dispatch loop crashes, integration failures.
- **debug** — high-volume internal state changes; always pair with a `module` field so it can be filtered.
Pass structured fields, not interpolated strings — fields are what makes a log queryable:
```ts
// Yes
logger.warn({ session_id, state }, "rejected append against non-live session");
// No
logger.warn(`session ${session_id} (${state}) rejected append`);
```
If you add a new typed error or a new event topic, also log it. The typed error tells the immediate caller; the event tells subscribers; the log tells the operator. The three are independent surfaces for the same fact, and all three are needed — events don't show up in the log, logs don't show up on the bus, errors don't survive the catch.
#### Structure
- Keep modules small. If a file is over ~500 lines, ask whether it has more than one responsibility.
- Prefer explicit composition (caller wires things together) over implicit dependency-injection magic.
- One public concept per module. If a file exports both `Foo` and `unrelatedHelperUsedNowhereElse`, the helper has the wrong home.
#### Dependencies
**Always check the registry for the current version before adding or updating a dependency.** Don't pick versions from memory.
The simple test: did you actually run `<package-manager> view <pkg> version` (or its equivalent) before pinning? If not, you're guessing — and the guess is almost always behind, sometimes by a major version with active deprecation.
Why this matters:
- **Long-running codebases.** Stale dependency anchors compound. A v3 you picked because that's what was on your mind will still be v3 a year later, and the v4 migration gets harder every month it waits.
- **Active deprecations are real.** Common libraries are mid-deprecation cycles for older majors. Pinning to a soon-deprecated version on day one means the eventual upgrade is forced rather than chosen.
- **"Latest" doesn't mean "use without thinking."** Read the changelog for breaking changes; decide whether to take them now or pin behind explicitly. But the *default* posture should be: pick latest, and justify in writing if pinning behind.
If you must pin behind latest, leave a comment in the manifest (or a CHANGELOG note) explaining why — a known incompatibility with another dep, an in-progress migration, a tested-and-rejected major bump. *"I haven't checked"* is not an acceptable reason.
Apply to: adding a new dep, updating an existing one, scaffolding a new package. *Not* required for routine no-op installs that don't touch versions.
#### Doc-vs-spec discipline
Specification documents (the deeper ones that describe how a component must behave) are **functional specifications**, not narrative essays. The discipline:
- **Concrete data structures** — type definitions, schemas, DDL, JSON examples
- **Concrete function signatures** — input/output types, error types
- **Concrete event/message payloads** — name + schema + example
- **Concrete API signatures** — input/output schemas, side effects, permission requirements
- **Flow pseudocode** — when orchestration is non-obvious
- **Invariants** — explicitly listed: "after X happens, Y is always true"
- **Failure modes** — enumerated: "if X fails, the system does Y"
- **Acceptance tests** — statements another engineer can implement: "given input X, output Y"
Avoid in spec docs:
- Long prose explaining motivation (that goes in ADRs / decision records)
- Discussion of alternatives considered (also ADRs)
- Background context (README handles this)
- Marketing/aspirational tone
Spec docs are for implementation. Read one and you should know what to build.
### When in doubt
Ask the human user. Specifically:
- "I don't see a clean seam for this; the options I see are A or B. Which fits the architecture you have in mind?"
- "This requires a new contract / capability / interface. Should I draft it before implementing?"
- "I'm about to add a special case here; that smells like a missing abstraction. Do you want me to extract the abstraction or note the smell and defer?"
Don't ship band-aids and ask forgiveness later. The user has explicitly said: prevent that pattern.
-1
View File
@@ -1 +0,0 @@
AGENTS.md
-2
View File
@@ -63,8 +63,6 @@ services:
cd /app && node dist/server.js"
ports:
- "5000:5000"
volumes:
- ./local-seeders:/app/seeders
environment:
- PORT=5000
- REDIS_URL=redis://wwv-redis:6379
@@ -1,158 +0,0 @@
# ADR-0002: Source-of-truth for plugin identity and registration
## Status
Accepted
## Date
2026-05-20
## Context
Two related problems in the plugin system surfaced in the same period, and both have the same root cause: canonical plugin information was being derived in middleware translation layers (folder names, hard-coded lists) rather than at its proper source. Each problem is fixed by pushing the canonical truth back to its rightful owner and removing the static translation in the middle. Documenting them together makes the shared principle explicit and discourages future regressions of the same shape.
### Problem 1 — Engine seeder ID derived from folder name
The WorldWideView local data engine discovers seeders by scanning the filesystem under `/app/seeders`. Historically, the engine assigned each seeder an ID equal to its **folder name** (e.g., the folder `gpsjam/` became plugin ID `"gpsjam"`).
The frontend, however, identifies plugins by a separate **`id` field** declared in each plugin class (e.g., `"gps-jamming"`). The engine's `/manifest` endpoint listed available plugins using folder-derived IDs, and the frontend's `localEngineHasPlugin(pluginId)` performed an exact string match against that list.
This created a silent failure mode: six seeders had folder names that did not match their corresponding frontend plugin IDs:
| Folder name | Frontend plugin `id` |
|---|---|
| `gpsjam/` | `gps-jamming` |
| `sanctions/` | `international-sanctions` |
| `wildfires/` | `wildfire` |
| `civilUnrest/` | `civil-unrest` |
| `conflictEvents/` | `conflict-events` |
| `cyberAttacks/` | `cyber-attacks` |
When `localEngineHasPlugin('gps-jamming')` returned `false` (because the manifest listed `"gpsjam"`), the frontend silently fell through to the production cloud engine, bypassing the local engine entirely. The bug was invisible — no error, just incorrect routing.
A short-term patch introduced `SEEDER_ID_ALIASES` in `src/core/data/engineManifest.ts` — a frontend translation map from folder names to plugin IDs. This was explicitly a band-aid: it duplicated knowledge in a consumer rather than fixing the contract at the source, and would go stale as new seeders were added.
The engine's TypeScript source (`seeder-loader.ts`) also had a `toKebabCase()` helper intended to normalize camelCase folder names (e.g., `civilUnrest` → `civil-unrest`), but its compiled `dist/` was stale (the production image had not been rebuilt to include it), and `toKebabCase` alone cannot resolve abbreviation mismatches like `gpsjam` → `gps-jamming`.
### Problem 2 — Fresh-install plugin set hard-coded
The marketplace auto-seeder ([src/lib/marketplace/seedDefaultPlugins.ts](../../../src/lib/marketplace/seedDefaultPlugins.ts)) writes a starter set of plugins to the database on first boot so the globe isn't empty on day one. The list of plugins to seed lived in a hard-coded constant `DEFAULT_PLUGIN_IDS` ([src/lib/marketplace/defaultPlugins.ts](../../../src/lib/marketplace/defaultPlugins.ts)).
Two failure modes followed from the static list:
1. **Drift**: ten plugins were published to npm under `@worldwideview/*` and verified — surveillance-satellites, military-bases, embassies, seaports, lighthouses, spaceports, iranwarlive, undersea-cables, mineral-mines, nuclear-facilities — but never added to `DEFAULT_PLUGIN_IDS`. Fresh installs missed them silently.
2. **Phantom 404s**: two entries in the list (`fortiguard`, `nz-traffic-cameras`) were dev-only sandboxes in `local-plugins/`, never published to npm. Every fresh boot hit `https://marketplace.worldwideview.dev/api/plugins/fortiguard` → 404, logged a warning, and continued. The branch `fix/diagnose-plugins-404` was opened to diagnose those.
A separate, signed verified-plugins registry already existed at `marketplace.worldwideview.dev/api/registry`, loaded by [registryClient.ts](../../../src/lib/marketplace/registryClient.ts) with Ed25519 verification. It was used for trust-stamping installed plugins but not for selecting which plugins to seed. That registry is the authoritative list of plugins that have been reviewed and verified — exactly the list a starter set should come from.
### Shared shape
Both problems wrap the same anti-pattern: a consumer (frontend, seeder) maintains a static translation/list derived from the producer's actual identity (seeder export, npm publication). The translation drifts. The fix in both cases is the same: stop translating in the consumer; let the producer declare the truth and consume it directly.
## Decisions
### Decision 1 — The seeder's exported `name` field is the canonical plugin ID
The engine reads each seeder's `dist/index.mjs` default export's `name` property and uses it as the plugin ID everywhere:
- `/manifest` endpoint `plugins` array
- WebSocket welcome message plugin list
- WebSocket data message `pluginId` field
- Redis key (`data:<name>:live`)
- Scheduler logs
The folder name is used only as a filesystem fallback if the seeder exports no `name`. The engine logs a warning at startup whenever the kebab-cased folder name differs from the seeder's exported `name`, making drift visible without blocking startup.
**Contract:** Every seeder MUST export a `name` field that exactly equals the corresponding frontend plugin's `id` field (kebab-case). This is a two-place contract: declare the same ID string in the seeder export and in the frontend plugin class.
Implementation: `seeder-loader.ts` was updated to set `id = seederConfig.name || toKebabCase(folderName)` instead of `id = toKebabCase(folderName)`. The frontend `SEEDER_ID_ALIASES` map was removed entirely. No translation layer exists between the engine's manifest and the frontend's plugin lookup.
### Decision 2 — The signed verified registry is the source-of-truth for fresh-install auto-seeding
The marketplace auto-seeder iterates the set returned by `getVerifiedPluginIds()` (the signed registry) instead of a hard-coded `DEFAULT_PLUGIN_IDS` constant. Publishing a plugin to the verified registry automatically makes it part of the fresh-install starter set on subsequent boots, with no code change.
The static `DEFAULT_PLUGIN_IDS` list is deleted. There is no fallback list in code — if the registry is unreachable on first attempt, the seeder defers without marking-seeded so the next request retries (the 5-min in-memory cache in `registryClient.ts` bounds the request rate during outages).
Trust stamping is collapsed: every plugin landed by the seeder is `verified` by definition, so the per-plugin `verified.has(pluginId)` check inside the seeder is dropped — `manifest.trust = "verified"` is set directly.
**Contract:** A plugin is part of the fresh-install starter set iff it appears in the signed verified-plugins registry. There is no other list to maintain.
## Shared principle
> **Canonical plugin truth lives at the source. Reject static translation/lookup layers in the middle.**
This is the architectural rule both decisions encode. Concrete corollaries for future contributions:
- Don't introduce a static map in a consumer to compensate for a producer that mis-labels itself. Fix the producer.
- Don't introduce a static list in code to enumerate what a registry or filesystem already enumerates. Read from the registry/filesystem.
- A translation layer that "just needs an entry added each time we publish a plugin" is a foot-gun — it will silently drift the moment someone forgets.
## Alternatives Considered
### For Decision 1
#### Folder renames (rejected as primary fix)
Rename the six mismatched folders to match their frontend plugin IDs (e.g., `gpsjam/` → `gps-jamming/`). This would make the folder name the canonical ID.
- **Rejected as primary fix** because Windows Docker Desktop holds NTFS volume mount locks on running containers, making renames impossible without a full Docker Desktop shutdown. More importantly, it makes the canonical ID fragile — any accidental folder rename breaks routing silently.
- Folder renames remain a cosmetic cleanup option, now that the seeder export is the actual source of truth.
#### Frontend alias map (`SEEDER_ID_ALIASES`)
Maintain a static mapping in `engineManifest.ts` that translates engine-reported IDs to frontend plugin IDs.
- **Rejected** because it puts translation responsibility on the consumer rather than the producer. Every new seeder with a naming mismatch requires a manual alias entry. The map goes stale silently — no warning if an alias is wrong or missing.
#### `toKebabCase()` normalization only
Apply `toKebabCase()` to folder names in the engine and rely on the convention that folder names encode the plugin ID in camelCase.
- **Rejected** because `toKebabCase` cannot resolve abbreviation mismatches (`gpsjam` → still `gpsjam`, not `gps-jamming`) or singular/plural differences (`wildfires` → still `wildfires`, not `wildfire`). It only handles camelCase-to-kebab conversion.
#### Strict validation with engine startup failure
Refuse to load any seeder whose folder name doesn't match its exported `name`, blocking startup until inconsistencies are resolved.
- **Rejected** as too aggressive for a local dev environment. A warning is sufficient — startup continues, mismatches are visible, and the system keeps working.
### For Decision 2
#### Add the missing 10 IDs to `DEFAULT_PLUGIN_IDS` (rejected)
Extend the static list to include the newly-published plugins.
- **Rejected** because it ships a fix that resolves the immediate symptom while preserving the underlying drift mechanism. The list would go stale again the next time a plugin is published.
#### Hybrid: registry as default + static fallback list
Read from the registry; fall back to a hard-coded list if the registry is empty/unreachable.
- **Rejected** because it reintroduces the static list as a permanent shadow source-of-truth. Fallback lists go stale silently. The chosen "defer seeding on empty registry" behavior is safer — fresh installs retry on the next request rather than locking in a stale set.
#### Delete auto-seeding entirely (rejected)
Drop the seeder and let users install everything from the marketplace UI starting from an empty globe.
- **Rejected** because fresh installs should have something to look at on day one. The user explicitly stated this requirement: "we want the user to have something to get started with."
## Consequences
### Easier
**Decision 1 — engine seeder identity**
- Self-documenting identity. Reading any seeder's `src/index.ts` immediately tells you its plugin ID. No external registry to consult.
- Filesystem layout is irrelevant. Folders can be named for readability or organized in subdirectories without affecting routing.
- No translation layers. The frontend receives the correct ID directly from the manifest; `localEngineHasPlugin()` works with a direct string comparison.
- Drift is visible. The engine warns on startup when folder names diverge from exported names, catching mistakes during development rather than silently at runtime.
- Adding a new seeder requires only: (1) export `name: '<plugin-id>'` matching the frontend plugin's `id`, (2) no registration step, no alias map entry.
**Decision 2 — marketplace seeding**
- Publishing a plugin to the verified registry is sufficient to add it to the starter set. No code change, no PR to add an ID to a constant.
- Phantom 404s are gone. The seeder only queries IDs that exist in the verified registry, which by construction reflects what's actually published.
- Trust semantics align. "Verified" and "auto-seeded on fresh install" are now the same condition, not two near-overlapping lists that can drift.
- Registry outages no longer brick fresh installs. The deferral semantics ensure the seeder retries instead of marking-seeded with zero rows.
### Harder
**Decision 1 — engine seeder identity**
- Seeder author bears responsibility. If a seeder exports the wrong `name`, routing fails silently (same failure mode as before, but now localized to the seeder rather than the folder). The startup warning mitigates this.
- Contract validation is advisory, not enforced. There is no build-time check that the seeder's `name` matches an actual frontend plugin `id`. A validation script (`scripts/validate-seeder-ids.mjs`) is the recommended mitigation.
- Engine image must be kept up to date. The fix lives in `dist/seeder-loader.js` inside the Docker image. If the published `ghcr.io/silvertakana/wwv-data-engine:latest` image is behind the local source, the behavior reverts to folder-name-based IDs. Local changes require `docker cp` injection or an image rebuild until the fix is published.
**Decision 2 — marketplace seeding**
- The verified registry must be kept accurate. A newly-published plugin doesn't auto-seed until it's added to the registry and the registry is re-signed. Operational responsibility shifts from "remember to add the ID to a code constant" to "remember to add the ID to the registry" — a similar discipline, but now at the right layer (the marketplace, not the consumer).
- Existing instances are not back-filled. Users whose `Setting` table already has `defaults_seeded=true` will not automatically receive newly-verified plugins; they must install via the marketplace UI. This is by design — re-seeding could reinstall plugins a user explicitly uninstalled.
- Dev-only sandboxes (`local-plugins/<name>/`) no longer "tag along" via the default list. `fortiguard` and `nz-traffic-cameras` only reach users once published to npm and added to the verified registry. The local-plugins folder discovery in `/api/marketplace/load` still surfaces them in dev (`NODE_ENV=development` or `WWV_PLUGIN_DEV=true`).
+13 -13
View File
@@ -1,6 +1,6 @@
{
"name": "worldwideview",
"version": "2.16.2",
"version": "2.15.15",
"private": true,
"type": "module",
"packageManager": "[email protected]",
@@ -31,8 +31,8 @@
"@prisma/adapter-pg": "^7.8.0",
"@prisma/client": "^7.8.0",
"@prisma/client-runtime-utils": "7.8.0",
"@sentry/nextjs": "^10.53.1",
"@supabase/supabase-js": "^2.106.0",
"@sentry/nextjs": "^10.52.0",
"@supabase/supabase-js": "^2.105.4",
"@types/pg": "^8.20.0",
"@vercel/analytics": "^1.6.1",
"@worldwideview/wwv-plugin-sdk": "workspace:*",
@@ -42,20 +42,20 @@
"hls.js": "^1.6.16",
"insecam-api": "^0.0.3",
"jose": "^6.2.3",
"lucide-react": "^1.16.0",
"lucide-react": "^1.14.0",
"next": "16.2.6",
"next-auth": "^5.0.0-beta.30",
"openid-client": "^6.8.4",
"pg": "^8.21.0",
"pg": "^8.20.0",
"prisma": "^7.8.0",
"react": "19.2.6",
"react-dom": "19.2.6",
"react-player": "^3.4.0",
"resium": "1.21.1",
"satellite.js": "^7.0.1",
"resium": "1.21.0",
"satellite.js": "^7.0.0",
"stripe": "^22.1.1",
"tslib": "^2.8.1",
"undici": "^8.3.0",
"undici": "^8.2.0",
"ws": "^8.20.1",
"zod": "^4.4.3",
"zustand": "^5.0.13"
@@ -74,9 +74,9 @@
"@types/react": "^19",
"@types/react-dom": "^19",
"@types/ws": "^8.18.1",
"@typescript-eslint/eslint-plugin": "^8.59.4",
"@typescript-eslint/parser": "^8.59.4",
"@vitejs/plugin-react": "^6.0.2",
"@typescript-eslint/eslint-plugin": "^8.59.3",
"@typescript-eslint/parser": "^8.59.3",
"@vitejs/plugin-react": "^5.1.4",
"@vitest/coverage-v8": "4.1.6",
"concurrently": "^9.2.1",
"copy-webpack-plugin": "^14.0.0",
@@ -92,9 +92,9 @@
"jsdom": "^28.1.0",
"rollup-plugin-external-globals": "^0.13.0",
"tree-kill": "^1.2.2",
"tsx": "^4.22.2",
"tsx": "^4.21.0",
"typescript": "^5",
"vite": "^8.0.13",
"vite": "^8.0.12",
"vitest": "^4.1.6"
},
"pnpm": {
+2 -2
View File
@@ -13,11 +13,11 @@
"clean": "rm -rf dist"
},
"dependencies": {
"commander": "^14.0.3",
"commander": "^12.1.0",
"prompts": "^2.4.2"
},
"devDependencies": {
"@types/node": "^25.9.1",
"@types/node": "^20.0.0",
"@types/prompts": "^2.4.9",
"typescript": "^5.0.0"
}
+2 -8
View File
@@ -41,15 +41,9 @@ export default defineConfig({
},
{
name: 'firefox',
use: {
use: {
...devices['Desktop Firefox'],
storageState: 'playwright/.auth/user.json',
launchOptions: {
firefoxUserPrefs: {
// Disable MSAA/AT-SPI accessibility layer — causes hangs on Linux CI
'accessibility.force_disabled': 1,
},
},
storageState: 'playwright/.auth/user.json'
},
},
{
+657 -905
View File
File diff suppressed because it is too large. Load diff
+23 -219
View File
@@ -7,7 +7,6 @@
*/
import fs from "fs";
import os from "os";
import path from "path";
import { build } from "vite";
import { fileURLToPath } from "url";
@@ -27,72 +26,9 @@ const EXTERNAL_GLOBALS = {
"resium": "globalThis.__WWV_HOST__.Resium",
};
/**
* Synthetic host-module exports for `@/` imports that plugins use to access
* the host application's runtime objects (store, plugin manager, components).
*
* When a plugin sub-component does `import { useStore } from "@/core/state/store"`,
* the tsconfig `@/*` alias (inherited from the main app root tsconfig) would normally
* resolve that to the actual `src/core/state/store.ts` file — pulling the entire Zustand
* store (and its transitive deps like `src/core/edition.ts`) into the bundle.
* This plugin intercepts those imports at the resolveId stage, before Vite touches
* the filesystem, and substitutes a synthetic module that reads from the host globals
* injected at runtime by `hostGlobals.ts`.
*/
const HOST_MODULE_SYNTHETICS = {
"@/core/state/store": `export const useStore = globalThis.__WWV_HOST__.useStore;`,
"@/core/plugins/PluginManager": `export const pluginManager = globalThis.__WWV_HOST__.pluginManager;`,
"@/components/video/CameraStream": `export const CameraStream = globalThis.__WWV_HOST__.CameraStream;`,
"zustand": `
const zustand = globalThis.__WWV_HOST__.zustand || {};
export default zustand;
export const { create, createStore } = zustand;
`,
};
/**
* @function wwvHostRedirectPlugin
* @description Vite plugin that intercepts host-app module imports (`@/core/...`,
* `@/components/...`) in local plugin builds and redirects them to the runtime
* host globals instead of bundling the full app source tree.
*
* @returns {import('vite').Plugin}
*/
function wwvHostRedirectPlugin() {
return {
name: "wwv-host-redirect",
enforce: "pre",
resolveId(id) {
if (id in HOST_MODULE_SYNTHETICS) return `\0${id}`;
return null;
},
load(id) {
if (!id.startsWith("\0")) return null;
const originalId = id.slice(1);
return HOST_MODULE_SYNTHETICS[originalId] ?? null;
},
};
}
/**
* @function readJsonFile
* @description Reads a UTF-8 JSON file, tolerating a leading BOM (U+FEFF).
* Node's `fs.readFileSync(p, 'utf-8')` preserves the BOM byte, which `JSON.parse`
* then rejects with "Unexpected token". Scaffolders and editors on Windows
* occasionally emit BOMs into `package.json`; pnpm/npm strip them silently,
* so we match that behavior here to avoid breaking dev on otherwise-valid files.
* @param {string} filePath
* @returns {any}
*/
function readJsonFile(filePath) {
let text = fs.readFileSync(filePath, "utf-8");
if (text.charCodeAt(0) === 0xFEFF) text = text.slice(1);
return JSON.parse(text);
}
/**
* @function discoverLocalPlugins
* @description Finds all directories in `local-plugins/` that contain a valid
* @description Finds all directories in `local-plugins/` that contain a valid
* `package.json` with a `worldwideview` manifest block.
* @returns {Array<{dir: string, manifest: any, pluginDir: string}>}
*/
@@ -104,141 +40,43 @@ export function discoverLocalPlugins() {
if (dir.startsWith(".")) return false;
const pkgPath = path.join(LOCAL_PLUGINS_DIR, dir, "package.json");
if (!fs.existsSync(pkgPath)) return false;
const pkg = readJsonFile(pkgPath);
const pkg = JSON.parse(fs.readFileSync(pkgPath, "utf-8"));
return !!pkg.worldwideview;
})
.map(dir => {
const pkgPath = path.join(LOCAL_PLUGINS_DIR, dir, "package.json");
const pkg = readJsonFile(pkgPath);
const pkg = JSON.parse(fs.readFileSync(pkgPath, "utf-8"));
const manifest = pkg.worldwideview;
manifest.version = pkg.version;
manifest.name = pkg.name;
manifest.description = pkg.description;
return { dir, manifest, pkg, pluginDir: path.join(LOCAL_PLUGINS_DIR, dir) };
return { dir, manifest, pluginDir: path.join(LOCAL_PLUGINS_DIR, dir) };
});
}
/**
* @function resolvePluginEntry
* @description Resolves the TypeScript source entry file for a plugin.
*
* Resolution order:
* 1. `worldwideview.dev_entry` — explicit override in the manifest block
* 2. `pkg.types` — if it points to a .ts/.tsx source file (not a .d.ts declaration)
* 3. `pkg.source` — explicit source field (less common, but valid)
* 4. Heuristic fallback: src/index.ts → src/index.tsx → index.ts → index.tsx
*
* @param {string} pluginDir - Absolute path to the plugin directory.
* @param {any} manifest - The parsed `worldwideview` manifest block.
* @param {any} pkg - The full parsed package.json object.
* @returns {string|null} Absolute path to the entry file, or null if not found.
*/
function resolvePluginEntry(pluginDir, manifest, pkg) {
const candidates = [];
if (manifest.dev_entry) {
candidates.push(path.join(pluginDir, manifest.dev_entry));
}
// pkg.types points to the TS source in all WWV plugins (e.g. "src/index.ts")
if (pkg.types && !pkg.types.endsWith(".d.ts")) {
candidates.push(path.join(pluginDir, pkg.types));
}
if (pkg.source) {
candidates.push(path.join(pluginDir, pkg.source));
}
// Heuristic fallbacks
for (const rel of ["src/index.ts", "src/index.tsx", "index.ts", "index.tsx"]) {
candidates.push(path.join(pluginDir, rel));
}
return candidates.find(f => fs.existsSync(f)) ?? null;
}
const SKIP_DIRS = new Set(["dist", "node_modules", ".git", ".turbo", ".vite"]);
/**
* @function latestSourceMtime
* @description Walks a plugin directory and returns the newest mtime (ms) across
* source files. Used to decide whether a cached `dist/frontend.mjs` is still valid.
* Skips build/output dirs so this is fast even on cold disks.
* @param {string} rootDir
* @returns {number} newest mtime in ms, or 0 if no files found
*/
function latestSourceMtime(rootDir) {
let newest = 0;
/** @param {string} dir */
function walk(dir) {
let entries;
try {
entries = fs.readdirSync(dir, { withFileTypes: true });
} catch {
return;
}
for (const entry of entries) {
if (entry.name.startsWith(".") && entry.name !== ".env") continue;
if (SKIP_DIRS.has(entry.name)) continue;
const full = path.join(dir, entry.name);
if (entry.isDirectory()) {
walk(full);
} else if (entry.isFile()) {
const m = fs.statSync(full).mtimeMs;
if (m > newest) newest = m;
}
}
}
walk(rootDir);
return newest;
}
/**
* @function isBuildFresh
* @description Returns true if the plugin's existing `dist/frontend.mjs` is newer
* than every source file under its directory — meaning Vite would produce a
* byte-identical bundle and the build can be skipped.
* @param {string} pluginDir
* @returns {boolean}
*/
function isBuildFresh(pluginDir) {
const distFile = path.join(pluginDir, "dist", "frontend.mjs");
if (!fs.existsSync(distFile)) return false;
const distMtime = fs.statSync(distFile).mtimeMs;
const sourceMtime = latestSourceMtime(pluginDir);
return distMtime >= sourceMtime;
}
/**
* @function buildPlugin
* @description Invokes the Vite build engine to compile a plugin's source
* @description Invokes the Vite build engine to compile a plugin's source
* into a single ES module, externalizing shared host dependencies.
*
* Returns `"fresh"` if an up-to-date `dist/frontend.mjs` already exists (no build run),
* `true` if Vite produced a new bundle, or `false` on failure.
*/
export async function buildPlugin({ dir, manifest, pkg, pluginDir }) {
if (isBuildFresh(pluginDir)) {
return "fresh";
}
export async function buildPlugin({ dir, manifest, pluginDir }) {
const devEntry = manifest.dev_entry || "index.ts";
let entryFile = path.join(pluginDir, devEntry);
const entryFile = resolvePluginEntry(pluginDir, manifest, pkg ?? {});
if (!entryFile) {
console.warn(`[sync] ⚠ No entry file found for ${dir}, skipping`);
return false;
// Fallback: check for .tsx
if (!fs.existsSync(entryFile)) {
const tsxEntry = devEntry.replace(".ts", ".tsx");
entryFile = path.join(pluginDir, tsxEntry);
if (!fs.existsSync(entryFile)) {
console.warn(`[sync] ⚠ No entry file found for ${dir}, skipping`);
return false;
}
}
try {
await build({
root: pluginDir,
// Disable auto-loading the plugin's own vite.config.ts — the local
// plugins' node_modules may be symlinked to a different repo's SDK
// build (worktree scenario), which breaks config loading. The sync
// script provides all necessary config inline.
configFile: false,
logLevel: "warn",
plugins: [wwvHostRedirectPlugin()],
build: {
lib: {
entry: entryFile,
@@ -276,7 +114,7 @@ export async function buildPlugin({ dir, manifest, pkg, pluginDir }) {
}
}
export function syncToPublic({ dir, manifest, pluginDir }, { quiet = false } = {}) {
export function syncToPublic({ dir, manifest, pluginDir }) {
const publicName = manifest.id || dir.replace("wwv-plugin-", "");
const targetDir = path.join(OUTPUT_DIR, publicName);
const distFile = path.join(pluginDir, "dist", "frontend.mjs");
@@ -313,7 +151,7 @@ export function syncToPublic({ dir, manifest, pluginDir }, { quiet = false } = {
JSON.stringify(pluginJson, null, 2)
);
if (!quiet) console.log(`[sync] ✅ ${publicName} → public/plugins-local/${publicName}/`);
console.log(`[sync] ✅ ${publicName} → public/plugins-local/${publicName}/`);
}
// Clean stale plugins from public/plugins-local/ that no longer exist in local-plugins/
@@ -328,32 +166,6 @@ function cleanStale(activeIds) {
}
}
/**
* @function runWithConcurrency
* @description Runs an async worker over `items` with a fixed concurrency cap.
* Vite builds are CPU-bound; running all ~30 in parallel thrashes the scheduler
* and balloons RSS. A cap around half the logical cores keeps wall-clock low
* without starving the rest of the dev environment (Next.js, the data engine).
* @template T, R
* @param {T[]} items
* @param {number} concurrency
* @param {(item: T) => Promise<R>} worker
* @returns {Promise<R[]>}
*/
async function runWithConcurrency(items, concurrency, worker) {
const results = new Array(items.length);
let cursor = 0;
const workers = Array.from({ length: Math.min(concurrency, items.length) }, async () => {
while (true) {
const i = cursor++;
if (i >= items.length) return;
results[i] = await worker(items[i]);
}
});
await Promise.all(workers);
return results;
}
export async function syncAll() {
const plugins = discoverLocalPlugins();
@@ -363,20 +175,12 @@ export async function syncAll() {
return;
}
const concurrency = Math.max(2, Math.min(plugins.length, Math.ceil(os.cpus().length / 2)));
console.log(`[sync] Found ${plugins.length} local plugin(s); building with concurrency=${concurrency}`);
const started = Date.now();
console.log(`[sync] Found ${plugins.length} local plugin(s): ${plugins.map(p => p.dir).join(", ")}`);
let freshCount = 0;
let builtCount = 0;
await runWithConcurrency(plugins, concurrency, async (plugin) => {
const result = await buildPlugin(plugin);
if (result === "fresh") freshCount++;
else if (result) builtCount++;
if (result) syncToPublic(plugin, { quiet: result === "fresh" });
});
console.log(`[sync] Done in ${((Date.now() - started) / 1000).toFixed(1)}s (${builtCount} built, ${freshCount} cached)`);
for (const plugin of plugins) {
const ok = await buildPlugin(plugin);
if (ok) syncToPublic(plugin);
}
const activeIds = plugins.map(p => p.manifest.id || p.dir.replace("wwv-plugin-", ""));
cleanStale(activeIds);
+10 -36
View File
@@ -15,66 +15,40 @@ if (!fs.existsSync(LOCAL_PLUGINS_DIR)) {
let debounceTimeout = null;
let isSyncing = false;
let pendingSync = false;
/** Files changed since last sync completed — collected during the debounce window */
const pendingChanges = new Set();
function handleFileChange(eventType, filename) {
if (!filename) return;
if (
filename.includes("dist") ||
filename.includes("node_modules") ||
if (filename && (
filename.includes("dist") ||
filename.includes("node_modules") ||
filename.endsWith(".map") ||
filename.includes(".git")
) return; // ignore build artifacts and git internals
pendingChanges.add(filename);
)) return; // ignore build artifacts and git internals
if (debounceTimeout) {
clearTimeout(debounceTimeout);
}
debounceTimeout = setTimeout(async () => {
if (isSyncing) {
pendingSync = true;
return;
}
await runSync();
await runSync(filename);
}, 500);
}
/**
* Pulls plugin slugs out of changed file paths so we can name what triggered the rebuild.
* fs.watch on `local-plugins/` emits paths like `aviation/src/index.ts` (POSIX or Windows).
*/
function summarizeChangedPlugins(files) {
const slugs = new Set();
for (const f of files) {
const head = f.split(/[\\/]/)[0];
if (head) slugs.add(head);
}
return [...slugs];
}
async function runSync() {
async function runSync(filename) {
isSyncing = true;
const changed = summarizeChangedPlugins(pendingChanges);
pendingChanges.clear();
const label = changed.length === 0
? "pending changes"
: changed.length <= 3
? changed.join(", ")
: `${changed.slice(0, 3).join(", ")} (+${changed.length - 3} more)`;
console.log(`\n[watch] 🔄 Change detected in: ${label}`);
console.log(`\n[watch] Change detected in local plugins (file: ${filename}). Syncing...`);
try {
await syncAll();
console.log(`[watch] ✨ Hot-reload ready`);
} catch (err) {
console.error(`[watch] Sync failed:`, err);
} finally {
isSyncing = false;
if (pendingSync) {
pendingSync = false;
await runSync();
await runSync("pending changes");
}
}
}
+4 -16
View File
@@ -129,27 +129,15 @@ export function AppShell() {
console.log("[AppShell] Platform Ready. Waiting for globe tiles...");
};
// Guard so boot only starts once regardless of which trigger fires first.
let bootStarted = false;
const startBootOnce = (reason: string) => {
if (bootStarted) return;
bootStarted = true;
console.log(`[AppShell] Starting boot sequence (${reason}).`);
// Start boot sequence when globe tiles are loaded
const unsubGlobe = dataBus.on("globeReady", () => {
console.log("[AppShell] Globe ready — starting boot sequence.");
boot.startBoot();
};
// Primary trigger: globe signals it is ready.
const unsubGlobe = dataBus.on("globeReady", () => startBootOnce("globeReady"));
// Safety fallback: if the globe never initialises (e.g. WebGL unavailable
// in headless CI environments), force the boot sequence after 20 s so the
// app still reaches the "ready" state and tests are not left hanging.
const safetyTimer = setTimeout(() => startBootOnce("safety-timeout"), 20_000);
});
startPlatform();
return () => {
clearTimeout(safetyTimer);
unsubGlobe();
boot.cleanup();
pluginManager.destroy();
+1 -1
View File
@@ -64,7 +64,7 @@ describe("resolveEngineUrl", () => {
await fetchLocalEngineManifest();
const url = resolveEngineUrl("plugin-local");
expect(url).toContain("localhost:5000/stream");
expect(url).toContain("localhost:5001/stream");
});
it("should fall back to plugin's custom streamUrl if not local", () => {
+1 -3
View File
@@ -36,8 +36,6 @@ describe("WsClient", () => {
beforeEach(() => {
vi.useFakeTimers();
vi.clearAllMocks();
// Pin Math.random to 0 so jitter is deterministic (delay = base only)
vi.spyOn(Math, "random").mockReturnValue(0);
// Mock WebSocket global
mockWs = {
@@ -87,7 +85,7 @@ describe("WsClient", () => {
);
});
it("should attempt to reconnect after base delay if the connection is closed while subscriptions exist", () => {
it("should attempt to reconnect after 5s if the connection is closed while subscriptions exist", () => {
wsClient.subscribe("plugin-a", "ws://engine-1/stream");
mockWs.onclose();
+8 -43
View File
@@ -9,23 +9,11 @@ interface EngineConnection {
subscriptions: Set<string>;
/** Grace period timer — closes the connection if no plugins remain subscribed */
cleanupTimer: NodeJS.Timeout | null;
/** Backoff attempt counter — resets after a stable connection (>5s open) */
reconnectAttempts: number;
/** Timer that resets the backoff counter once a connection has been stable */
stableConnectionTimer: NodeJS.Timeout | null;
}
const RECONNECT_BASE_MS = 5000;
const RECONNECT_MAX_MS = 60000; // Cap at 1 minute
const RECONNECT_JITTER_MS = 4000;
const STABLE_CONNECTION_MS = 5000; // Reset backoff after 5s of stable connection
const RECONNECT_DELAY_MS = 5000;
const CLEANUP_GRACE_MS = 30000;
/** Normalizes underscore-based pluginIds to kebab-case (e.g. `my_plugin` → `my-plugin`). */
function normalizePluginId(id: string): string {
return id.replace(/_/g, "-");
}
class WebSocketClient {
private engines = new Map<string, EngineConnection>();
@@ -37,8 +25,6 @@ class WebSocketClient {
reconnectTimer: null,
subscriptions: new Set(),
cleanupTimer: null,
reconnectAttempts: 0,
stableConnectionTimer: null,
};
this.engines.set(engineUrl, engine);
}
@@ -57,12 +43,6 @@ class WebSocketClient {
engine.ws.onopen = () => {
console.debug(`[WSClient] 🟢 Connected to ${engineUrl}. WS Handshake took ${(performance.now() - wsStart).toFixed(2)}ms`);
// Only reset backoff if the connection stays open for a non-trivial time —
// an immediate close (e.g. server-side rejection) shouldn't be treated as success.
if (engine.stableConnectionTimer) clearTimeout(engine.stableConnectionTimer);
engine.stableConnectionTimer = setTimeout(() => {
engine.reconnectAttempts = 0;
}, STABLE_CONNECTION_MS);
// Resubscribe to all active plugins on this engine
for (const pluginId of engine.subscriptions) {
this.send(engine, { action: "subscribe", pluginId });
@@ -94,39 +74,25 @@ class WebSocketClient {
};
engine.ws.onclose = () => {
console.warn(`[WSClient] Disconnected from ${engineUrl}. Reconnecting in 5s...`);
engine.ws = null;
if (engine.stableConnectionTimer) {
clearTimeout(engine.stableConnectionTimer);
engine.stableConnectionTimer = null;
}
if (engine.reconnectTimer) clearTimeout(engine.reconnectTimer);
// Only reconnect if there are still active subscriptions
if (engine.subscriptions.size > 0) {
// Exponential backoff with jitter to prevent thundering herd on engine restart.
// 5s -> 10s -> 20s -> 40s -> 60s (cap), plus ±4s of jitter so simultaneous
// sessions don't all reconnect at the same instant.
const expDelay = Math.min(
RECONNECT_BASE_MS * Math.pow(2, engine.reconnectAttempts),
RECONNECT_MAX_MS
);
const delay = expDelay + Math.random() * RECONNECT_JITTER_MS;
engine.reconnectAttempts++;
console.warn(`[WSClient] Disconnected from ${engineUrl}. Reconnecting in ${Math.round(delay / 1000)}s (attempt ${engine.reconnectAttempts})...`);
engine.reconnectTimer = setTimeout(() => this.connectEngine(engineUrl), delay);
engine.reconnectTimer = setTimeout(() => this.connectEngine(engineUrl), RECONNECT_DELAY_MS);
}
};
}
private handleDataMessage(data: WsStreamPayload) {
const pluginId = normalizePluginId(data.pluginId!);
const plugin = pluginManager.getPlugin(pluginId)?.plugin;
const plugin = pluginManager.getPlugin(data.pluginId!)?.plugin;
let finalEntities = data.payload as GeoEntity[];
const existingEntities = useStore.getState().entitiesByPlugin[pluginId] || [];
const existingEntities = useStore.getState().entitiesByPlugin[data.pluginId!] || [];
if (plugin && typeof (plugin as any).mapWebsocketPayload === "function") {
finalEntities = (plugin as any).mapWebsocketPayload(data.payload, existingEntities);
} else if (!Array.isArray(data.payload)) {
console.warn(`[WsClient] Payload for ${pluginId} is an object but no mapWebsocketPayload exists. Ignoring.`);
console.warn(`[WsClient] Payload for ${data.pluginId} is an object but no mapWebsocketPayload exists. Ignoring.`);
return;
} else {
finalEntities = finalEntities.map((e) => ({
@@ -135,10 +101,10 @@ class WebSocketClient {
}));
}
console.debug(`[WSClient] 🔄 Dispatching ${finalEntities.length} entities for ${pluginId} to DataBus`);
console.debug(`[WSClient] 🔄 Dispatching ${finalEntities.length} entities for ${data.pluginId} to DataBus`);
dataBus.emit("dataUpdated", {
pluginId,
pluginId: data.pluginId!,
entities: finalEntities,
});
}
@@ -177,7 +143,6 @@ class WebSocketClient {
if (engine.subscriptions.size === 0) {
console.log(`[WSClient] No subscriptions remain for ${engineUrl}. Closing connection.`);
if (engine.reconnectTimer) clearTimeout(engine.reconnectTimer);
if (engine.stableConnectionTimer) clearTimeout(engine.stableConnectionTimer);
engine.ws?.close();
this.engines.delete(engineUrl);
}
+25 -12
View File
@@ -8,24 +8,37 @@ let manifestFetched = false;
/**
* Resolve the base URL of the local data engine.
*
* Always checks localhost:5000 — the port docker-compose.yml binds for
* wwv-data-engine. NEXT_PUBLIC_WWV_PLUGIN_DATA_ENGINE_URL is intentionally
* NOT used here: that variable belongs to each plugin's own declared engine
* URL (production, third-party, etc.) and must not poison local detection.
* Mixing the two caused the production engine to be reported as "local".
* Priority:
* 1. `NEXT_PUBLIC_WWV_PLUGIN_DATA_ENGINE_URL` env var — if the operator
* pointed plugins at a specific engine URL (most common in
* self-host setups), use that. We strip any trailing `/stream`
* since the var is sometimes set to the WebSocket URL.
* 2. `localhost:5001` — matches the port wwv-data-engine's
* docker-compose.yaml binds. The previous default of 5000 was
* always wrong: the engine has listened on 5001 since the v2
* refactor. Port 5000 is also famously hijacked on macOS by the
* AirPlay Receiver, which would return 403 to any GET regardless
* of whether an engine was running, so it produced misleading
* "no local engine" results on every macOS dev machine.
*/
function getLocalEngineBase() {
if (typeof window === "undefined") return "http://localhost:5000";
return `${window.location.protocol}//${window.location.hostname}:5000`;
const envUrl = process.env.NEXT_PUBLIC_WWV_PLUGIN_DATA_ENGINE_URL;
if (envUrl) {
// Strip a trailing `/stream` if present (the var is sometimes
// set to the WebSocket URL); also normalize ws[s]:// to http[s]://.
return envUrl
.replace(/\/stream\/?$/, "")
.replace(/^ws:\/\//, "http://")
.replace(/^wss:\/\//, "https://")
.replace(/\/+$/, "");
}
if (typeof window === "undefined") return "http://localhost:5001";
return `${window.location.protocol}//${window.location.hostname}:5001`;
}
/**
* Fetch the list of available seeders from a local engine.
* Returns null if no local engine is detected (timeout after 500ms).
*
* The engine guarantees manifest IDs are already in kebab-case (the seeder's
* exported `name` field is the canonical plugin ID). No client-side translation
* is needed — what the engine reports is what the frontend uses.
* Returns null if no local engine is detected (timeout after 2s).
*/
export async function fetchLocalEngineManifest(): Promise<string[] | null> {
if (manifestFetched) return localManifest;
+5 -2
View File
@@ -20,8 +20,11 @@ function toWsStreamUrl(url: string): string {
const DEFAULT_ENGINE_URL = toWsStreamUrl(RAW_ENGINE_URL);
function getLocalWsUrl() {
if (typeof window === "undefined") return "ws://localhost:5000/stream";
return `${window.location.protocol === "https:" ? "wss:" : "ws:"}//${window.location.hostname}:5000/stream`;
if (process.env.NEXT_PUBLIC_WWV_PLUGIN_DATA_ENGINE_URL) {
return DEFAULT_ENGINE_URL;
}
if (typeof window === "undefined") return "ws://localhost:5001/stream";
return `${window.location.protocol === "https:" ? "wss:" : "ws:"}//${window.location.hostname}:5001/stream`;
}
/**
* Resolves the WebSocket engine URL for a given plugin.
+31
View File
@@ -0,0 +1,31 @@
/**
* Marketplace plugins automatically installed on fresh instances.
*
* These provide a rich out-of-the-box experience — the globe looks alive
* from the very first load. Users can uninstall any of these afterwards.
*
* All core plugins are now installed via marketplace defaults.
*/
export const DEFAULT_PLUGIN_IDS = [
"aviation",
"maritime",
"military-aviation",
"wildfire",
"camera",
"borders",
"osm-search",
"earthquakes",
"satellite",
"daynight",
"conflict-zones",
"volcanoes",
"airports",
"international-sanctions",
"gps-jamming",
"fortiguard",
"nz-traffic-cameras",
"air-defense",
"civil-unrest",
"conflict-events",
"cyber-attacks",
] as const;
+25 -36
View File
@@ -4,6 +4,7 @@ import {
} from "vitest";
import { seedDefaultPlugins } from "./seedDefaultPlugins";
import { DEFAULT_PLUGIN_IDS } from "./defaultPlugins";
// ---------------------------------------------------------------------------
// Mocks — must be declared before importing the module under test
@@ -49,9 +50,6 @@ vi.stubGlobal("fetch", mockFetch);
// Helpers
// ---------------------------------------------------------------------------
/** Fake verified registry payload used by most tests. */
const FAKE_VERIFIED_IDS = ["aviation", "maritime", "surveillance-satellites"];
function fakeManifest(id: string) {
return {
id,
@@ -74,11 +72,10 @@ describe("seedDefaultPlugins", () => {
mockIsDemo = false;
delete process.env.WWV_SKIP_DEFAULT_PLUGINS;
// Defaults: not seeded, no existing plugins, manifests valid,
// verified registry returns the fake set.
// Defaults: not seeded, no existing plugins, manifests valid
mockFindFirst.mockResolvedValue(null);
mockCount.mockResolvedValue(0);
mockGetVerifiedPluginIds.mockResolvedValue(new Set(FAKE_VERIFIED_IDS));
mockGetVerifiedPluginIds.mockResolvedValue(new Set(DEFAULT_PLUGIN_IDS));
mockValidateManifest.mockReturnValue({ valid: true, errors: [] });
mockUpsertPlugin.mockResolvedValue({});
mockCreate.mockResolvedValue({});
@@ -93,11 +90,11 @@ describe("seedDefaultPlugins", () => {
}));
});
it("seeds every plugin in the verified registry on a fresh install", async () => {
it("seeds all default plugins on a fresh install", async () => {
await seedDefaultPlugins();
expect(mockUpsertPlugin).toHaveBeenCalledTimes(FAKE_VERIFIED_IDS.length);
for (const id of FAKE_VERIFIED_IDS) {
expect(mockUpsertPlugin).toHaveBeenCalledTimes(DEFAULT_PLUGIN_IDS.length);
for (const id of DEFAULT_PLUGIN_IDS) {
expect(mockUpsertPlugin).toHaveBeenCalledWith(
id,
"1.0.0",
@@ -120,7 +117,6 @@ describe("seedDefaultPlugins", () => {
expect(mockUpsertPlugin).not.toHaveBeenCalled();
expect(mockFetch).not.toHaveBeenCalled();
expect(mockGetVerifiedPluginIds).not.toHaveBeenCalled();
});
it("marks seeded without installing when plugins already exist", async () => {
@@ -137,28 +133,14 @@ describe("seedDefaultPlugins", () => {
);
});
it("does not markSeeded when the verified registry is empty", async () => {
// Simulates registry outage / signature failure / no cache
mockGetVerifiedPluginIds.mockResolvedValue(new Set<string>());
await seedDefaultPlugins();
expect(mockUpsertPlugin).not.toHaveBeenCalled();
expect(mockFetch).not.toHaveBeenCalled();
// Critical: guard row is NOT written so the next request retries
expect(mockCreate).not.toHaveBeenCalled();
expect(mockUpdateMany).not.toHaveBeenCalled();
});
it("handles marketplace manifest fetch failure gracefully", async () => {
it("handles marketplace API failure gracefully", async () => {
mockFetch.mockRejectedValue(new Error("Network error"));
await seedDefaultPlugins();
// No upserts (every manifest fetch failed), but guard row IS written
// — we got a non-empty verified set, so seeding completed (with zero
// successes); next request shouldn't retry the whole flow.
// Should NOT throw
expect(mockUpsertPlugin).not.toHaveBeenCalled();
// Guard row still written (fail-safe)
expect(mockCreate).toHaveBeenCalled();
});
@@ -169,7 +151,6 @@ describe("seedDefaultPlugins", () => {
expect(mockUpsertPlugin).not.toHaveBeenCalled();
expect(mockFetch).not.toHaveBeenCalled();
expect(mockGetVerifiedPluginIds).not.toHaveBeenCalled();
// Guard row written so it doesn't re-check
expect(mockCreate).toHaveBeenCalled();
});
@@ -181,7 +162,6 @@ describe("seedDefaultPlugins", () => {
expect(mockUpsertPlugin).not.toHaveBeenCalled();
expect(mockFetch).not.toHaveBeenCalled();
expect(mockGetVerifiedPluginIds).not.toHaveBeenCalled();
expect(mockCreate).not.toHaveBeenCalled();
});
@@ -197,16 +177,25 @@ describe("seedDefaultPlugins", () => {
await seedDefaultPlugins();
// One fewer than total (the first was skipped)
expect(mockUpsertPlugin).toHaveBeenCalledTimes(FAKE_VERIFIED_IDS.length - 1);
expect(mockUpsertPlugin).toHaveBeenCalledTimes(DEFAULT_PLUGIN_IDS.length - 1);
});
it("stamps every seeded plugin as verified", async () => {
it("stamps trust from the verified registry", async () => {
const firstId = DEFAULT_PLUGIN_IDS[0];
// Only the first plugin is verified
mockGetVerifiedPluginIds.mockResolvedValue(new Set([firstId]));
await seedDefaultPlugins();
expect(mockUpsertPlugin).toHaveBeenCalledTimes(FAKE_VERIFIED_IDS.length);
for (const call of mockUpsertPlugin.mock.calls) {
const manifest = JSON.parse(call[2]);
expect(manifest.trust).toBe("verified");
}
// Check the first call has "verified"
const firstCall = mockUpsertPlugin.mock.calls[0];
expect(firstCall[0]).toBe(firstId);
const firstManifest = JSON.parse(firstCall[2]);
expect(firstManifest.trust).toBe("verified");
// Check a non-verified one
const secondCall = mockUpsertPlugin.mock.calls[1];
const secondManifest = JSON.parse(secondCall[2]);
expect(secondManifest.trust).toBe("unverified");
});
});
+17 -26
View File
@@ -1,6 +1,8 @@
/* eslint-disable no-console */
import { isDemo } from "@/core/edition";
import { validateManifest } from "@/core/plugins/validateManifest";
import { prisma } from "../db";
import { DEFAULT_PLUGIN_IDS } from "./defaultPlugins";
import { upsertPlugin } from "./repository";
import { getVerifiedPluginIds } from "./registryClient";
@@ -8,20 +10,16 @@ const MARKETPLACE_URL = process.env.NEXT_PUBLIC_MARKETPLACE_URL
|| "https://marketplace.worldwideview.dev";
/**
* Seed verified marketplace plugins on a fresh install.
* Seed default marketplace plugins on a fresh install.
*
* The signed registry (`getVerifiedPluginIds`) is the single source of truth
* for which plugins land in a brand-new instance — there is no hard-coded
* default list. Publish a plugin to the verified registry and it auto-seeds
* on subsequent fresh installs.
* Runs once per instance lifecycle — an idempotent guard
* (`defaults_seeded` in the Setting table) prevents re-runs.
*
* Runs at most once per instance lifecycle: an idempotent guard
* (`defaults_seeded` in the Setting table) prevents re-runs. If the registry
* is empty or unreachable on first attempt, the guard is NOT set — the next
* request retries.
* Like the "sample data" that ships with a new app: the seeder
* writes records to the database on first boot, then never runs again.
*
* Errors are logged but never thrown — a failed seed must never block the
* application from starting.
* Errors are logged but never thrown — a failed seed must never
* block the application from starting.
*/
export async function seedDefaultPlugins(): Promise<void> {
try {
@@ -47,29 +45,22 @@ export async function seedDefaultPlugins(): Promise<void> {
return;
}
const verified = await getVerifiedPluginIds();
if (verified.size === 0) {
// Registry unreachable / signature failed / empty — defer so the
// next request retries instead of locking in an empty fresh install.
console.warn(
"[DefaultPlugins] Verified registry returned empty — deferring seed, will retry next request",
);
return;
}
console.log(
`[DefaultPlugins] Fresh install detected — seeding ${verified.size} verified plugins…`,
`[DefaultPlugins] Fresh install detected — seeding ${DEFAULT_PLUGIN_IDS.length} default plugins…`,
);
const verified = await getVerifiedPluginIds();
let installed = 0;
for (const pluginId of verified) {
for (const pluginId of DEFAULT_PLUGIN_IDS) {
try {
const manifest = await fetchManifest(pluginId);
if (!manifest) continue;
// Every plugin in the verified set is by definition verified.
manifest.trust = "verified";
// Server-side trust stamping
manifest.trust = verified.has(pluginId)
? "verified"
: "unverified";
// Reconstruct CDN entry for npm-distributed plugins
if (manifest.npmPackage) {
@@ -102,7 +93,7 @@ export async function seedDefaultPlugins(): Promise<void> {
await markSeeded();
console.log(
`[DefaultPlugins] Seeded ${installed}/${verified.size} plugins`,
`[DefaultPlugins] Seeded ${installed}/${DEFAULT_PLUGIN_IDS.length} plugins`,
);
} catch (err) {
console.error("[DefaultPlugins] Seeder failed:", err);