observability

Router skill. Resolve topic, then Read EXACTLY ONE playbook from routes:. [DOC]

When to use

Production-health work: stand up monitoring, design alerts, automate health checks, manage logs, run an incident. Not app perf-tuning or CI/CD. [INFERENCIA]

Inputs → Outputs

• In: topic (required), depth (quick|deep), target system/stack context.
• Out: routed playbook applied to the user's system — config, runbook, or remediation, with evidence tags on every non-obvious claim. [DOC]

Resolve the topic

Pick the single best match; route by intent, not keyword overlap. [INFERENCIA]

• monitoring-setup — first-time metrics/dashboards/SLO instrumentation.
• alerting-strategy — thresholds, routing, noise/fatigue, on-call paging.
• health-check-automation — liveness/readiness probes, synthetic checks.
• log-management — structured logging, aggregation, retention, querying.
• incident-response — live outage: triage, comms, mitigation, postmortem.

Disambiguate: "alert too noisy" → alerting-strategy (not monitoring-setup); "is it up?" automation → health-check-automation; "what broke just now?" → incident-response. If still ambiguous after one read of the request, ASK — never guess past a {VACIO_CRITICO}. [DOC]

Spine, gates & done-when

Discover → Analyze → Execute → Validate. deep = apply playbook exhaustively with verification each step; quick = essentials only. Enforce constitution v6.0.0, Alfa evidence tags, the script-first rule. [CONFIG] Done when: routed topic matches intent; exactly one playbook loaded; output tagged; deep runs verified each step. [DOC]

Anti-patterns

Loading the whole cluster; routing on a single keyword; auto-filling a missing target system instead of asking; emitting untagged claims. [INFERENCIA]

Assets

Routing and acceptance policies live in assets/ (routing-matrix.json, quality-rubric.json); see assets/README.md. [CONFIG]