self-observe

name	self-observe
description	Observe openppx runtime health with token usage, gateway/heartbeat status, error logs, and quick diagnostics.

Self Observe Skill

Use this skill when the user asks for agent self-inspection, runtime health checks, or diagnostics such as token cost, error logs, and service status.

What To Check

1. Token usage:

ppx token stats --json
ppx token stats --provider google --limit 50 --json
ppx token stats --provider openai --limit 50 --json

2. Runtime status:

ppx gateway status --json
ppx heartbeat status --json
ppx cron status
ppx provider status --json

3. Error logs (read-only):

tail -n 200 ~/.openppx/log/gateway.err.log
rg -n "ERROR|Error|Traceback|Exception|failed|timeout" ~/.openppx/log/gateway.err.log ~/.openppx/log/gateway.out.log ~/.openppx/log/gateway.debug.log

4. SQLite quick verification (if sqlite3 exists):

sqlite3 ~/.openppx/token_usage.db "SELECT provider, COUNT(*) AS requests, SUM(total_tokens) AS total_tokens FROM llm_token_usage_events GROUP BY provider ORDER BY total_tokens DESC;"
sqlite3 ~/.openppx/token_usage.db "SELECT response_at, provider, model, request_tokens, response_tokens, total_tokens FROM llm_token_usage_events ORDER BY response_at_ms DESC LIMIT 20;"

Fast Path

Generate one consolidated report:

bash openppx/skills/self-observe/scripts/self_status_report.sh

Output Format

When reporting to user, include:

1. Runtime Summary: gateway/heartbeat/provider/cron highlights.
2. Token Summary: total requests/tokens and provider split.
3. Recent Errors: latest error signatures with file and timestamp.
4. Risks: what might break soon (missing usage data, repeated failures, disconnected provider).
5. Next Actions: concrete commands to validate/fix.

Guardrails

• Keep checks read-only by default.
• Do not delete or truncate logs unless the user explicitly asks.
• If status/log files are missing, report "not found" explicitly instead of guessing.
• Prefer structured output (--json) first, then summarize in natural language.