AgentSkillsCN

Observe Triage

观察与分类

SKILL.md

Skill: Observe Triage

Repeatable workflow for the Observe stage: classify a signal quickly and route it to a concrete next action (fix now, schedule, or document).

When To Use

  • A user reports broken behavior.
  • Output logs show repeated errors.
  • WS bridge/browser dev mode fails to connect.
  • Automations appear stuck or not scheduling.

Inputs

  • Symptom description (what happened)
  • Where it happened (route/surface)
  • Environment (VS Code webview vs browser dev via WS bridge)
  • Any logs/screenshots

Outputs

  • Classification + likely root cause
  • Minimal repro steps (if possible)
  • Next step:
    • Fix now (patch + verification)
    • Plan it (new task + owner)
    • Document it (docs/runbook update)

Workflow

  1. Open the canonical runbook

    • runbooks/observe-triage.md
  2. Confirm environment + transport

    • VS Code webview transport: webview.postMessage
    • Browser dev transport: WS bridge ws://127.0.0.1:5174/?token=...
    • Use System → Registry to confirm:
      • deep link base (scheme)
      • WS endpoint + token presence
      • universal routes/actions present
  3. Classify quickly

    • Auth/config (baseUrl/email/token)
    • Transport (IPC vs WS bridge, CSP/connect-src, token mismatch)
    • Data (Jira API/JQL/network)
    • Routing (deep link parsing, missing route, redirects)
  4. Route to the right fix path

    • If fix now: implement smallest patch + validate (bun run typecheck)
    • If schedule: create a Plan-stage next action + owner
    • If document: update _agents/docs/* or _agents/runbooks/*

Safety Rules

  • Prefer diagnosis + repro before code changes.
  • Confirm before changing user settings or secrets.
  • Confirm before creating issues in external systems.

Related

  • Runbook: runbooks/observe-triage.md
  • Docs: docs/routing-matrix.md, docs/configuration-matrix.md, docs/universal-matrix.md