| name | feedback-observing |
| description | Read agent interaction logs across packs and detect operational patterns. Emits state-transition FeedbackEvents for degradation/recovery. |
| user-invocable | true |
| allowed-tools | Read, Write, Glob, Grep, Bash |
Feedback Observing
Read agent interaction logs across all packs, detect operational patterns (error rates, duration outliers, zero-invocation skills), and emit state-transition FeedbackEvents when patterns change.
This skill closes the agent feedback loop — agents report back what worked and what failed.
Core Principle
State-transition only emission. Only emit FeedbackEvents when a skill's operational state changes (healthy → concerning, concerning → healthy). This prevents recursive signal inflation from repeated analysis of the same stable state.
Triggers
/feedback-observe # Analyze all packs' agent logs
/feedback-observe --pack observer # Specific pack only
/feedback-observe --since 7d # Time range (default: 30d)
When to Use
- Periodically to check agent health across packs
- After deploying new skills or modifying existing ones
- When suspecting a skill is silently failing
- As part of a daily/weekly maintenance routine
Workflow
Phase 1: Collect Logs
Glob: grimoires/*/agent-logs/*.jsonl
For each JSONL file:
Parse each line (skip malformed lines, log warning)
Filter by --pack if specified
Filter by --since if specified (compare ts field)
Collect all log entries into a unified dataset.