Add CLAUDE.md with repo guidance

[jainankit]

Summary

• Adds a CLAUDE.md at the repo root to orient future Claude Code sessions.
• Covers commands (uv install, celerymon local run, mypy, pre-commit, Docker build) and notes the absence of a test suite.
• Documents the three-watcher / one-collector architecture, the intentional custom UUID cache in EventWatcher, the auto-reconnect behavior, and /healthz semantics.
• Captures conventions worth preserving: metric sub-namespacing, create_started factory pattern, and SPDX headers.

Test plan

• No code changes; verify rendering of CLAUDE.md on the PR page.

[aviator-app]

Current Aviator status

Aviator will automatically update this comment as the status of the PR changes.
Comment /aviator refresh to force Aviator to re-examine your PR (or learn about other /aviator commands).

This PR was merged using Aviator.

---

See the real-time status of this PR on the Aviator webapp.

Use the Aviator Chrome Extension to see the status of your PR within GitHub.

[gemini-code-assist]

Code Review

This pull request introduces CLAUDE.md to provide architectural guidance and development instructions for the celerymon project. A review comment pointed out that the documentation's description of histogram accumulation reveals a double-accumulation bug in the existing codebase and suggested updating the file to accurately reflect the current implementation and data structures.

[jainankit]

/aviator revise

[aviator-app]

No runbook found for this PR. Creating one — this may take a moment... View Runbook

[aviator-app]

Review comments addressed successfully!

Summary of changes:

• Line 26 (EventWatcher sentence): Dropped the incorrect sentence about histogram buckets being stored as "parallel defaultdict[task_name, int] lists accumulated into Prometheus cumulative form at collection time in metrics.py". The actual data structure is task_runtime_sec keyed by (task_name, result) and accumulation happens in the watcher itself.

• Line 28 (metrics.py paragraph): Replaced the blanket "there is no locking, watchers update fields atomically by reassignment" claim with an accurate description: RedisWatcher/WorkerWatcher use atomic reassignment, while EventWatcher mutates accumulators in-place so collect() takes defensive snapshots before iterating.