davidtoby Agent Platform Operations
Use this umbrella for operational work on Hermes/OpenClaw/Clawdbot-like agent platforms. If the current task is specifically Hermes Agent setup/configuration, also load the official Hermes skill required by the system.
Core workflow
- Identify platform, checkout path, install method, gateway mode, and current command/error.
- Prefer documented CLIs and config files over ad-hoc edits.
- Before destructive changes, create backups of config/state directories.
- Reproduce with logs enabled; capture exact command, exit code, and relevant log excerpt.
- Apply the smallest fix; then verify by running the affected CLI/gateway/tool path.
- Report changed files, commands run, and rollback path.
Labeled playbooks
Manual update and report
Use when the normal updater hangs or is blocked by approval. Pull/update the checkout, inspect status, run a smoke test, and produce a concise maintenance report.
OpenClaw CLI/docs/cheatsheet
Use docs search and the cheatsheet for exact subcommands and flags; do not invent CLI syntax.
Backups and restore
Archive the platform state directory with exclusions for caches/logs/secrets as appropriate. Verify archive creation and restore instructions.
Cost auditing
Parse local usage logs or provider metrics, summarize spend by model/provider/time, and highlight runaway loops or high-cost patterns.
Memory documentation and self-improvement
Capture stable, reusable lessons in memory/skills; avoid storing transient task progress.
Shield/safety hardening
Treat safety policy/config changes as high-risk: inspect current config, make explicit diffs, and verify behavior.
Feishu approval and websocket stability
For button/callback failures, provide fallback approval paths and inspect callback sync. For websocket reconnect delays or duplicate gateways, check app_id collisions, ping intervals, and running processes.
Reference files
Historical platform-specific recovery recipes and exact errors are stored under references/from-*.md.