Imported from Kevt-t/job-ops-personalv (AGENTS.md). Install upstream with npx skills add Kevt-t/job-ops-personalv. Copyright stays with the author.

Error/Logging/Sanitization Standards

This project uses strict operability and privacy defaults for server-side code.

API Response Contract

For all /api/* routes, return:

• Success: { ok: true, data, meta?: { requestId } }
• Error: { ok: false, error: { code, message, details? }, meta: { requestId } }

Use consistent status/code mapping:

• 400 INVALID_REQUEST
• 401 UNAUTHORIZED
• 403 FORBIDDEN
• 404 NOT_FOUND
• 408 REQUEST_TIMEOUT
• 409 CONFLICT
• 422 UNPROCESSABLE_ENTITY
• 500 INTERNAL_ERROR
• 502 UPSTREAM_ERROR
• 503 SERVICE_UNAVAILABLE

Correlation IDs

• Honor inbound x-request-id when present; otherwise generate one.
• Always return x-request-id header.
• Include request ID in API responses (meta.requestId) and logs.
• Propagate context into async flows (especially pipeline run and per-job work) so logs include pipelineRunId / jobId when available.

Logging Rules

• Use the shared logger wrapper (infra/logger.ts) in core server paths.
• Do not add direct console.log, console.warn, or console.error in core paths.
• Log structured objects, not free-form dumps.
• Include useful context fields (e.g. requestId, pipelineRunId, jobId, route, status).

SSE Standards

• Use centralized SSE helpers by default.
• Server: use orchestrator/src/server/infra/sse.ts for setup, data writes, comments, and heartbeats.
• Client (EventSource): use orchestrator/src/client/lib/sse.ts for subscription/open/message/error plumbing.
• Do not duplicate raw SSE setup (Content-Type, Connection, heartbeat loops, or ad-hoc JSON.parse event parsing) when these helpers apply.
• Keep feature payload types domain-local (pipeline, ghostwriter, bulk actions), but reuse shared transport plumbing.

Redaction and Sanitization

• Always sanitize objects before logging or returning in error details.
• Redact sensitive keys by default (authorization, cookie, password, secret, token, apiKey, etc.).
• Truncate large payloads and long strings.
• Do not throw/log raw upstream response bodies, full webhook bodies, or large JSON.stringify(...) blobs.

Webhook and LLM Payload Defaults

• Webhooks: send minimal whitelisted payloads by default.
• LLM prompts: send only required profile/job fields; avoid unnecessary PII.
• Document external payload behavior when adding new integrations.

PR Checklist (Routes/Services)

• API responses follow { ok, data/error, meta.requestId }.
• Status/code mapping is correct and consistent.
• Request/correlation IDs appear in logs and async workflows.
• No raw sensitive payload logging or raw upstream body throws.
• New/changed webhook or LLM payloads are sanitized and documented.

Documentation Standards (Condensed)

When adding or updating user-facing docs:

• Use this feature-page structure:
  1. What it is
  2. Why it exists
  3. How to use it
  4. Common problems
  5. Related pages
• Include frontmatter keys: id, title, description, sidebar_position.
• Prefer concrete, step-by-step instructions over abstract explanation.
• Include copy-pasteable examples where relevant.
• State defaults and constraints explicitly.
• Link related docs with /docs/... URLs.
• Any user-visible behavior change should include corresponding docs updates.

Validation / Verification

Before marking work complete, verify changes with the same checks used by CI.

Required CI-parity checks

Run from repository root:

1. ./orchestrator/node_modules/.bin/biome ci .
2. npm run check:types:shared
3. npm --workspace orchestrator run check:types
4. npm --workspace orchestrator run build:client
5. npm --workspace orchestrator run test:run

Native module note (better-sqlite3)

If tests fail with a Node ABI mismatch for better-sqlite3, rebuild it before running tests:

• npm --workspace orchestrator rebuild better-sqlite3

CI runs on Node 22. If local behavior differs, verify with Node 22 before concluding a change is valid.

Scope-specific checks

• For focused changes, run targeted tests first (for touched files/modules), then still run the full CI-parity list above before finalizing.
• A change is considered valid only when all required checks pass without ignored failures.