feat: improve error info for debugging (WS logs, /debug/processes, /api/status, docs)

[shingo257]

Summary

Improves error and debug information to make troubleshooting easier (wrangler tail, debug endpoints, and public status).

Changes

1. WebSocket proxy logging (src/index.ts)

   • Emit structured one-line logs for close and error with side, code, and reason/message so wrangler tail can be grepped easily.

2. /debug/processes (src/routes/debug.ts)

   • Query ?failed=1: return only gateway-related failed/completed processes (start-moltbot, clawdbot gateway).
   • When logs=true or failed=1, add lastFailedStderrPreview: sanitized stderr (first 500 chars) from the most recent failed start-moltbot process.
   • New helper src/utils/sanitize.ts to redact secrets (e.g. sk-ant-*, Slack tokens) from log excerpts.

3. /api/status (src/routes/public.ts)

   • When the gateway process has failed or exited with non-zero code, fetch its stderr and expose a sanitized prefix as lastStderrPreview (300 chars) so callers can see a hint without calling /debug/processes.

4. Docs (AGENTS.md)

   • New Troubleshooting section: “Health OK but no reply” (check /api/status, /debug/processes?logs=true, wrangler tail for [WS] close/[WS] error), and exit code 126 (CRLF, tr -d '\r', chmod in Dockerfile).

Notes

• “Can’t automatically merge” is expected if cloudflare/moltworker main has new commits. You can still open this PR; maintainers can merge after resolving conflicts, or we can rebase this branch on upstream main and push again.

[andreasjansson]

This PR is based on the old architecture (references start-moltbot, clawdbot gateway). The startup script was rewritten in PR #186 (clawdbot → openclaw migration), WebSocket logging was improved in PR #133, and the persistence layer was replaced entirely in PR #329. The debugging improvements proposed here would need a complete rewrite to work with the current codebase. Thank you for the contribution.