Agent Skills: Skill: backend/observability-audit

Skill: backend/observability-audit

Purpose

Improve and standardize observability so the system is operable after changes:

• logs: structured, queryable, minimal but sufficient
• metrics: counters/timers/gauges for key outcomes
• traces: propagation and spans where available
• audit: records for sensitive actions when required

This skill is a pass over changed behavior, not feature work.

Inputs

• Changed components/endpoints/jobs
• Risk posture (from repo docs/profile if present)
• Observability standards (from REPO_PROFILE.json if present)
• Correlation/id propagation standard (if any)

Outputs

• Improved logging:
  • include correlation id
  • include primary identifiers (request id, user id where safe, entity id)
  • include outcome + error codes
• Metrics updates if repo uses them:
  • success/fail counters
  • latency timing
  • queue depth / retries (for jobs)
• Tracing spans or propagation fixes if repo uses tracing
• Audit entries for protected actions if applicable

Non-goals

• Implementing business logic
• Adding heavy instrumentation everywhere
• Logging secrets or sensitive payloads

Workflow

1. Identify repo observability standards (profile/docs).
2. Add structured logs at critical boundaries:
   • request start/end
   • job start/end
   • external call start/end (adapter boundary)
3. Ensure errors are observable:
   • log error codes/taxonomy, not raw stack spam only
   • include retry decisions (retrying vs terminal)
4. Add metrics if the repo uses them:
   • count outcomes
   • measure latency
5. Ensure correlation ids propagate:
   • inbound request → domain → adapter → logs
6. Add audit entries where required:
   • authz-protected actions
   • fund movement / order placement / credential updates (example categories)
7. Run validations.

Checks

• Logs answer: what happened, to what, by whom (when safe), and why
• No secrets logged; sensitive values redacted
• Correlation id present on critical flows (where available)
• Metrics/tracing updated when repo supports them
• Changes are minimal and focused on the modified behavior

Failure modes

• No logging standard → follow conservative structured logging and document assumptions.
• High-cardinality metrics risk → avoid unbounded labels.
• Sensitive data risk → redact or omit; prefer identifiers over payloads.

Telemetry

Log:

• skill: backend/observability-audit
• areas: logs | metrics | tracing | audit (subset)
• files_touched
• outcome: success | partial | blocked