arthai-marketplace

/sre

Run SRE operations.

Synopsis

/sre [command]

Commands: status, health, logs, deploy-check, ci, sleep, wake, debug, rebuild, watch. With no command, defaults to status. health accepts --monitor-mode (report only on failure — used by scheduled checks registered via watch).

When to use it

Checking whether your services and environments are up — before or after a deploy
Investigating a production or staging issue with a systematic, evidence-first approach
Day-to-day infra chores: viewing logs, sleeping/waking staging, clean monorepo rebuilds
Not for: fixing code bugs — that’s /fix (and /sre debug will route you there automatically); CI auto-remediation — that’s /ci-fix; “something is broken and I don’t know what” — start with /incident

Quickstart

/sre

What you’ll see: a structured pass/fail table covering every environment from your CLAUDE.md — deployment platform status, health endpoint results, and the latest CI runs.

Examples

/sre                          # full system status — all services, all environments
/sre health                   # quick check — just hit the health endpoints
/sre logs backend             # recent deploy logs for a service, errors summarized
/sre debug "500s on /api"     # systematic diagnosis of a production issue
/sre sleep                    # scale staging down to save costs
/sre rebuild                  # nuclear option for stale build caches

Subcommands

Command	What it does (one line)
`status`	Full system status — checks platform services, hits all health endpoints, and reviews recent CI runs for every environment
`health`	Quick health check — hits the health endpoint of every environment listed in CLAUDE.md and reports pass/fail
`logs [service]`	Fetches recent deploy logs for a service, filters for errors, and summarizes issues found
`deploy-check`	End-to-end deployment verification — recent CI runs, latest deploys, health endpoints, and migration state
`ci`	Checks CI/CD pipeline status, pulls failure logs for any failed runs, and suggests fixes
`sleep`	Sleeps the staging environment to save costs and verifies services scaled to zero
`wake`	Wakes the staging environment and waits for health checks to pass
`debug [description]`	Systematic production debugging — confirm, scope, diagnose, find root cause, then apply infra fixes directly or escalate code bugs to `/fix`
`rebuild`	Clean monorepo rebuild — stops services, nukes build caches, rebuilds packages in dependency order, restarts, and verifies
`watch [service]`	Registers a background health watcher via the Monitor tool — wakes the agent only on a status change; delegates to `/monitor deploy` when that skill is installed (falls back to a session-scoped scheduled check where Monitor is unavailable)

What it does

Discovers project context — reads CLAUDE.md, the project profile, and the knowledge base; detects your deploy platform, databases, monitoring, and CI from project files if not already calibrated, and recommends anything that’s missing.
Runs the command — most commands spawn an SRE agent with platform-appropriate tools (Railway MCP, Vercel CLI, gh, etc.); rebuild and watch run inline without an agent. Discovery is scoped per command — local-only commands read just the CLAUDE.md tables.
Classifies fixes during debug — infra-only fixes (restart, config, scale) are applied directly and health-verified; if config changed, it asks you to smoke-test — a user-confirmation checkpoint. Code bugs are never patched from /sre — they’re escalated to /fix with all gathered evidence.
Writes back what it learned — incidents, platform patterns, and infra decisions go to the project knowledge base so future runs diagnose faster.
After an infra fix, presents a “What’s next?” menu (monitor, investigate further, project status, done) — a user-confirmation checkpoint.

Agents spawned

Agent	Model tier	Role
sre	sonnet (from `model-policy.yml`)	Executes the command against your platform — status checks, log analysis, debugging, sleep/wake (one agent per invocation; `rebuild` and `watch` spawn none — they run inline)

Output & artifacts

Structured status/health reports in the conversation (pass/fail per environment)
Incident reports at .claude/qa-knowledge/incidents/<date>-<slug>.md after debug finds a root cause
Knowledge base appends: .claude/knowledge/agents/sre.md (platform patterns, resolutions) and .claude/knowledge/shared/decisions.md (infra decisions)
For code bugs: a documented handoff into /fix with root cause, affected files, and evidence

Troubleshooting

Problem	Fix
Environments are skipped during health checks	Their URL in the CLAUDE.md `## Environments` table is still a TODO placeholder — fill it in or run `/scan`
Platform not detected / wrong tools used	Run /calibrate — `/sre` reads its project profile instead of re-discovering
`/sre debug` found the bug but didn’t fix the code	By design — code fixes go through `/fix`, which has the verification pipeline. Accept the escalation it offers
Stale builds or phantom errors persist after restart	Use `/sre rebuild` — it clears turbo/next/tsc caches and rebuilds from scratch

/incident — the triage orchestrator; routes to /sre debug when the evidence points at infrastructure
/restart — local dev servers only; /sre covers deployed environments
/deploy — deploys to local/staging; /sre deploy-check verifies the result
/ci-fix — auto-remediation when /sre ci finds failures