evlog

Use this skill when work touches evlog — wide events, structured errors, drains, sampling, audit trails, AI SDK telemetry, client logging, or @evlog/cli map coverage scoring.

Workflow

1. Inspect the local logging surface before changing code:

- Package versions for evlog, optional @evlog/cli, framework peers, and drain destinations in use. - Integration path: Nuxt/Nitro module, Next createEvlog/withEvlog, TanStack Start evlog/nitro/v3, Hono/Express/Fastify/Elysia middleware, standalone initLogger, Workers, etc. - Patterns: scattered console/pino lines vs log.set wide events, createError({ why, fix }), audits, AI wraps, client ingest.

1. Refresh docs when versions or APIs are unclear. Start from [source-map.md](references/source-map.md).
2. For install, three logging modes, initLogger, and core APIs, use [setup-core.md](references/setup-core.md).
3. For wide-event design, sealing/fork, structured errors, catalogs, and redaction, use [wide-events-errors.md](references/wide-events-errors.md).
4. For framework wiring and adapter choice, use [frameworks-adapters.md](references/frameworks-adapters.md).
5. For pipelines, sampling, audit, AI, client logs, and evlog map/CI, use [cli-advanced.md](references/cli-advanced.md).

Implementation Judgment

• Prefer one wide event per unit of work (log.set then auto-emit or manual emit) over many info lines in the same request/job.
• Group context into nested objects (user, cart, payment) with meaningful keys—not flat userId/cartId soup or a generic data bag.
• Use the right mode: log.* for one-shots; createLogger / createRequestLogger when you own the lifecycle; framework useLogger / req.log / c.get('log') when middleware owns create+emit.
• Throw createError({ message, status, why, fix, link?, code?, internal? }) instead of bare Error. Parse on clients with parseError.
• After emit (or sampled drop), the logger is sealed—post-response set is ignored. Use log.fork (where available) for intentional background child events.
• Libraries and shared packages: use createLogger / accept a logger; do not call initLogger and force a drain on the host.
• Production: wrap adapters in createDrainPipeline (batch + retry); enable head+tail sampling; keep errors and money/auth paths.
• On Nuxt/Nitro/Next/TanStack Start, run bunx @evlog/cli map (or npx @evlog/cli map) before large logging refactors—fix FIX FIRST first. Pin CLI version in CI.

Verification

Prefer the repo's existing checks. For meaningful evlog changes, include the relevant subset:

• Hit a route/job locally and confirm one wide event with expected nested fields (pretty in dev).
• Force a createError path and confirm why/fix in logs and client parseError.
• bunx @evlog/cli map / doctor when changing entry-point coverage or setup.
• Drain smoke (Axiom/Sentry/fs NDJSON) when changing pipelines; flush on shutdown.
• Sampling config check: errors and keep rules still retained in production-shaped config.
• Typecheck for RequestLogger, typed fields, and catalog imports.