rockport-opslisted

## User Input ```text $ARGUMENTS ``` You **MUST** consider the user input before proceeding (if not empty). ## Purpose Diagnose infrastructure issues, fix them through the speckit pipeline (preserving full history), and produce structured advice for any calling Claude instance that depends on Rockport. Input can be: - Free-form symptom descriptions ("chat completions are timing out") - Raw error output pasted from another Claude Code instance - Health check requests ("is rockport healthy?") - Specific error messages or HTTP status codes ## Workflow Execute these 5 phases in order. Every phase is mandatory. ### Phase 1: Triage Investigate the issue using **subagents** to keep diagnostic noise out of the main context. The main context stays clean for reasoning and speckit. **Delegate diagnostic work to subagents.** Each subagent should: - Run the specific AWS CLI / SSM / curl commands needed - Return a **summary** (5-10 lines max), not raw output - Include specific error messages, timestamps, and status codes Use the diagnostic procedures in [diagnostics.md](references/diagnostics.md) for the exact commands. Work top-down through the layers, but skip layers that clearly aren't relevant to the symptom. **Layer order:** 1. Instance state (running? SSM reachable?) 2. Service health (systemd status for litellm, cloudflared, rockport-video, postgresql) 3. Health endpoints (localhost:4000/health, localhost:4001/v1/videos/health) 4. Recent logs (journalctl errors from the