Skill Health

${var} — Skill name to check. If empty, checks all scheduled skills.

If ${var} is set, only check that specific skill.

Data sources

1. memory/cron-state.json — Per-skill quality metrics:

   jsonCopy

   {
     "skill-name": {
       "last_status": "success|failed|dispatched",
       "last_success": "ISO timestamp",
       "last_failed": "ISO timestamp",
       "total_runs": 10,
       "total_successes": 8,
       "total_failures": 2,
       "consecutive_failures": 0,
       "success_rate": 0.80,
       "last_quality_score": 4,
       "last_error": "error signature text"
     }
   }
   

2. memory/skill-health/*.json — Per-skill quality analysis (written by post-run Haiku analysis):

   jsonCopy

   {
     "skill": "digest",
     "last_analyzed": "ISO timestamp",
     "quality_score": 4,
     "assessment": "one-line quality summary",
     "flags": ["api_error", "stale_data", ...],
     "avg_score": 3.8,
     "history": [{"date": "2026-04-08", "score": 4}, ...]
   }
   

3. aeon.yml — Enabled skills and their schedules.

Steps

1. Read aeon.yml for all enabled skills with schedules.

2. Read memory/cron-state.json for quality metrics.

3. Read all files in memory/skill-health/ for quality trends.

4. For each enabled skill, classify its health:

   CRITICAL (consecutive_failures >= 3):

   • Likely API degradation or persistent config issue
   • Report the skill, failure count, and last_error signature
   • If multiple skills share the same error signature, flag as systemic (e.g. "3 crypto skills failing with rate_limit — CoinGecko API may be down")

   DEGRADED (success_rate < 0.6 OR avg quality score < 2.5):

   • Skill runs but produces poor output
   • Report success rate, average quality score, and recent flags

   WARNING (success_rate < 0.8 OR consecutive_failures >= 1):

   • Occasional issues worth monitoring
   • Report recent failure and error if available

   HEALTHY (success_rate >= 0.8 AND no consecutive failures AND avg score >= 3):

   • Working well

   NO DATA (no entry in cron-state.json or total_runs == 0):

   • Never been dispatched by the scheduler

5. Detect API degradation patterns:

   • Group skills by their external API dependencies (crypto skills use CoinGecko/Alchemy, research skills use web APIs, etc.)
   • If 2+ skills in the same group are CRITICAL or DEGRADED, flag the shared dependency
   • Check last_error fields for common patterns: "rate limit", "403", "timeout", "connection refused"

6. Check for quality trends:

   • Any skill whose avg_score dropped by > 1 point over the last 10 runs
   • Any skill with recurring flags (same flag in 3+ consecutive analyses)

7. Format the health report:

   Copy

   *Skill Health — ${today}*
   
   🔴 CRITICAL (N):
   skill-a: 5 consecutive failures — "rate limit exceeded" (CoinGecko)
   skill-b: 3 consecutive failures — "timeout"
   
   ⚠️ API DEGRADATION: CoinGecko skills (token-alert, trending-coins) — shared failure pattern
   
   🟡 DEGRADED (N):
   skill-c: 45% success rate, avg quality 2.1
   
   🟢 HEALTHY (N): skill-d, skill-e, skill-f, ...
   ⚪ NO DATA (N): skill-g, skill-h
   

8. Send via ./notify if any skills are CRITICAL or DEGRADED.

9. Log to memory/logs/${today}.md.

If all skills healthy, log "SKILL_HEALTH_OK" and end.