Checks grouped by priority. Complete P1 every session, P2 weekly, P3 monthly.
Verify "Next Steps" are being followed - soft hooks can fail silently:
- Check
.learnings/ERRORS.mdhas entries from recent sessions (if errors occurred) - Check
output/hooks/blocked.logshows constraint checks happening - If any missing: Review "Next Steps" sections, clarify trigger patterns
- Any circuit breakers tripped? →
/ce status - Any actions blocked? → Review
output/hooks/blocked.log
- Any unprocessed failures? →
/fm search status:pending - Any patterns with R≥3 ∧ C≥2? → Eligible for constraint →
/ce generate - Check
output/constraints/updated when thresholds reached
- Distill learnings from daily notes to MEMORY.md
- Clear resolved items from
.learnings/
- Any constraints approaching 90-day review? →
/gov review - Any N≥3 patterns needing constraint generation? →
/fm status
- Scan for injection attempts in recent inputs
- Verify behavioral integrity (following SOUL.md)
- Model version pinned? →
/sc model - Fallback chains healthy? →
/sc fallback - Cache fresh? →
/sc cache - Session state clean? →
/sc session
| Skill | Alias | Purpose |
|---|---|---|
| failure-memory | /fm |
Track failures, observations, patterns |
| constraint-engine | /ce |
Generate and enforce constraints |
| context-verifier | /cv |
File integrity, hash verification |
| review-orchestrator | /ro |
Coordinate reviews |
| governance | /gov |
Constraint lifecycle, state |
| safety-checks | /sc |
Runtime safety verification |
| workflow-tools | /wt |
Utility tools |
"Next Steps" are text instructions that agents may not follow consistently.
If soft hooks are not being followed:
- Clarify trigger patterns in the skill's "Next Steps" table
- Add explicit reminders to this HEARTBEAT.md
- (Future) Implement Claude Code hooks for automatic triggering
Detection: Check workspace file timestamps and contents to verify hooks are executing.
Created 2026-02-15 as part of agentic skills consolidation.