agent-triagelisted

# /agent-triage — Smart Failure Triage > Failed tasks fall into a small number of buckets. Classify, then act. ## Prerequisites | Var | Purpose | |---|---| | `AGENT_API_URL` | Base URL of the agent's HTTP API | | `AGENT_API_TOKEN` | Bearer token | Endpoints used: - `GET /api/tasks?status=failed&since=<window>` — list failed tasks - `GET /api/tasks/:id/log` — full log for one task - `POST /api/tasks/:id/requeue` — requeue - `POST /api/tasks/:id/dead-letter` — dead-letter ## Workflow ### Phase 1: Pull failures ```bash curl -s -H "Authorization: Bearer $AGENT_API_TOKEN" \ "$AGENT_API_URL/api/tasks?status=failed&since=24h" ``` ### Phase 2: Classify each failure For each failed task, fetch its log/last error and assign one classification: | Class | Pattern | Recommended action | |---|---|---| | `infra` | API rate limit, credit exhausted, network timeout, 5xx from upstream | **requeue** — not the task's fault | | `auth` | 401/403 from upstream, expired token, missing credential | **fix-and-retry** — operator must rotate or refresh | | `code` | Compile error, test failure, lint failure that the agent could fix | **requeue** with prompt amendment, or **dead-letter** if 3+ retries already | | `prompt` | Agent looped, hit max iterations without converging, hallucinated nonexistent file | **fix-and-retry** — operator amends the prompt | | `env` | Missing dep, wrong Node version, path doesn't exist on worker | **fix-and-retry** — operator fixes the env | | `wontfix` | Task is