Flywheel Skill

Monitor the knowledge flywheel health.

The Flywheel Model

Sessions → Transcripts → Forge → Pool → Promote → Knowledge
     ↑                                               │
     └───────────────────────────────────────────────┘
                    Future sessions find it

Velocity = Rate of knowledge flowing through Friction = Bottlenecks slowing the flywheel

Execution Steps

Given /flywheel:

Step 1: Measure Knowledge Pools

# Count top-level artifact files (avoid counting directories)
LEARNINGS=$(find .agents/learnings -maxdepth 1 -type f 2>/dev/null | wc -l)
PATTERNS=$(find .agents/patterns -maxdepth 1 -type f 2>/dev/null | wc -l)
RESEARCH=$(find .agents/research -maxdepth 1 -type f 2>/dev/null | wc -l)
RETROS=$(find .agents/retros -maxdepth 1 -type f 2>/dev/null | wc -l)
echo "Learnings: $LEARNINGS"
echo "Patterns: $PATTERNS"
echo "Research: $RESEARCH"
echo "Retros: $RETROS"

Step 2: Check Recent Activity

# Recent learnings (last 7 days)
find .agents/learnings -maxdepth 1 -type f -mtime -7 2>/dev/null | wc -l
# Recent research
find .agents/research -maxdepth 1 -type f -mtime -7 2>/dev/null | wc -l

Step 3: Detect Staleness

# Old artifacts (> 30 days without modification)
find .agents/ -name "*.md" -mtime +30 2>/dev/null | wc -l

Step 3.5: Check Cache Health

if command -v ao &>/dev/null; then
  # Get citation report (cache metrics)
  CITE_REPORT=$(ao metrics cite-report --json --days 30 2>/dev/null)
  if [ -n "$CITE_REPORT" ]; then
    HIT_RATE=$(echo "$CITE_REPORT" | jq -r '.hit_rate // "unknown"')
    UNCITED=$(echo "$CITE_REPORT" | jq -r '(.uncited_learnings // []) | length')
    STALE_90D=$(echo "$CITE_REPORT" | jq -r '.staleness["90d"] // 0')
    echo "Cache hit rate: $HIT_RATE"
    echo "Uncited learnings: $UNCITED"
    echo "Stale (90d uncited): $STALE_90D"
  fi
else
  # ao-free fallback: compute approximate metrics from files
  echo "Cache health (ao-free fallback):"

  # Learnings modified in last 30 days (active pool)
  ACTIVE_30D=$(find .agents/learnings/ -name "*.md" -mtime -30 2>/dev/null | wc -l | tr -d ' ')
  echo "Active learnings (30d): $ACTIVE_30D"

  # Forge candidates awaiting promotion
  FORGE_PENDING=$(ls .agents/forge/*.md 2>/dev/null | wc -l | tr -d ' ')
  echo "Forge candidates pending: $FORGE_PENDING"

  # Citation tracking (if citations.jsonl exists)
  if [ -f .agents/ao/citations.jsonl ]; then
    CITATION_COUNT=$(wc -l < .agents/ao/citations.jsonl | tr -d ' ')
    UNIQUE_CITED=$(grep -o '"artifact_path":"[^"]*"' .agents/ao/citations.jsonl 2>/dev/null | sort -u | wc -l | tr -d ' ')
    echo "Total citations: $CITATION_COUNT"
    echo "Unique learnings cited: $UNIQUE_CITED"
  else
    echo "No citation data (citations.jsonl not found)"
  fi

  # Session outcomes (if outcomes.jsonl exists)
  if [ -f .agents/ao/outcomes.jsonl ]; then
    OUTCOME_COUNT=$(wc -l < .agents/ao/outcomes.jsonl | tr -d ' ')
    echo "Session outcomes recorded: $OUTCOME_COUNT"
  fi
fi

Step 4: Check ao CLI Status

if command -v ao &>/dev/null; then
  ao metrics flywheel status 2>/dev/null || echo "ao metrics flywheel status unavailable"
  ao status 2>/dev/null || echo "ao status unavailable"
  ao maturity --scan 2>/dev/null || echo "ao maturity unavailable"
  ao anti-patterns 2>/dev/null || echo "ao anti-patterns unavailable"
  ao badge 2>/dev/null || echo "ao badge unavailable"

  # Knowledge maintenance
  ao dedup --merge 2>/dev/null || true
  ao contradict 2>/dev/null || true
  ao constraint review 2>/dev/null || true
  ao curate status 2>/dev/null || true
  ao metrics health 2>/dev/null || true
  ao metrics cite-report --days 30 2>/dev/null || true

  # Active pruning: archive stale, evict low-utility, and curate noisy uncited learnings
  ao maturity --expire --archive 2>/dev/null || true
  ao maturity --evict --archive 2>/dev/null || true
  ao maturity --curate --archive 2>/dev/null || true

  # Retrieval quality: use the representative live corpus when it exists
  if [ -d cli/cmd/ao/testdata/retrieval-bench-live ]; then
    ao retrieval-bench --live --corpus cli/cmd/ao/testdata/retrieval-bench-live --json 2>/dev/null || true
  fi
else
  echo "ao CLI not available — using file-based metrics"

  # Pool inventory
  echo "Pool depths:"
  for pool in learnings patterns forge knowledge research retros; do
    COUNT=$(ls .agents/${pool}/*.md 2>/dev/null | wc -l | tr -d ' ')
    echo "  $pool: $COUNT"
  done

  # Global patterns
  GLOBAL_COUNT=$(ls ~/.claude/patterns/*.md 2>/dev/null | wc -l | tr -d ' ')
  echo "  global patterns: $GLOBAL_COUNT"

  # Check for promotion-ready learnings (see references/promotion-tiers.md)
  echo "See: skills/flywheel/references/promotion-tiers.md for tier definitions"
fi

Step 4.5: Process Metrics (from skill telemetry)

If .agents/ao/skill-telemetry.jsonl exists, use jq to extract: invocations by skill, average cycle time per skill, gate failure rates. Include in health report (Step 6) under ## Process Metrics.

Step 5: Validate Artifact Consistency

Cross-reference validation: scan knowledge artifacts for broken internal references. Use scripts/artifact-consistency.sh (method documented in references/artifact-consistency.md). Default allowlist lives at references/artifact-consistency-allowlist.txt; use --no-allowlist for a full raw audit.

Health indicator: >90% = Healthy, 70-90% = Warning, <70% = Critical.

Step 6: Write Health Report

Write to: .agents/flywheel-status.md

# Knowledge Flywheel Health

**Date:** YYYY-MM-DD

## Pool Depths
| Pool | Count | Recent (7d) |
|------|-------|-------------|
| Learnings | <count> | <count> |
| Patterns | <count> | <count> |
| Research | <count> | <count> |
| Retros | <count> | <count> |

## Velocity (Last 7 Days)
- Sessions with extractions: <count>
- New learnings: <count>
- New patterns: <count>

## Artifact Consistency
- References scanned: <count>
- Broken references: <count>
- Consistency score: <percentage>%
- Status: <Healthy/Warning/Critical>

## Cache Health
- Hit rate: <percentage>%
- Uncited learnings: <count>
- Stale (90d uncited): <count>
- Status: <Healthy/Warning/Critical>

## Retrieval Quality
- Live corpus coverage: <percentage or unavailable>
- Live corpus learnings: <count or unavailable>
- Status: <Healthy/Warning/Critical>

## Health Status
<Healthy/Warning/Critical>

## Friction Points
- <issue 1>
- <issue 2>

## Recommendations
1. <recommendation>
2. <recommendation>

Step 7: Report to User

Tell the user:

1. Overall flywheel health
2. Knowledge pool depths
3. Recent activity
4. Any friction points
5. Recommendations

Health Indicators

Metric	Healthy	Warning	Critical
Learnings/week	3+	1-2	0
Stale artifacts	<20%	20-50%	>50%
Research/plan ratio	>0.5	0.2-0.5	<0.2
Cache hit rate	>80%	50-80%	<50%

Golden Signals

Four golden signals (always shown) reveal whether knowledge is truly compounding or just accumulating noise.

ao flywheel status              # table output with golden signals
ao flywheel status --json       # machine-readable

The Four Signals

#	Signal	Question	Key Metric
1	Velocity Trend	Is σρ-δ increasing?	Linear regression slope of baseline velocities (7d/30d)
2	Citation Pipeline	Are citations useful?	% of feedback with reward > 0.6
3	Research Closure	Is research being mined?	% orphaned research (no learning backlink)
4	Reuse Concentration	Is the whole pool active?	Gini coefficient of citation distribution

Verdicts and Thresholds

Signal	Healthy	Warning	Critical
Velocity Trend	compounding (slope > +0.01)	stagnant	decaying (slope < -0.01)
Citation Pipeline	reinforcing (>60% high-util)	inert (30-60%)	degrading (<30%)
Research Closure	mining (<=10% orphans)	—	hoarding (>=10% orphans)
Reuse Concentration	distributed (Gini<0.4, active>30%)	concentrated	dormant (Gini>0.7 or active<10%)

Overall verdict: 3+ healthy = compounding, 3+ critical = decaying, mixed = accumulating.

Recommended Actions

Verdict	Action
decaying	Run /compile cycle, archive stale artifacts, increase citation via ao lookup
accumulating	Review orphaned research (/research→/retro pipeline), improve forge quality
compounding	Maintain cadence. Consider capturing baselines (ao metrics baseline) for trend tracking

Cache Eviction

Read references/cache-eviction.md for the full eviction pipeline (passive tracking → confidence decay → maturity scan → archive).

Key Rules

• Monitor regularly - flywheel needs attention; address bottlenecks early
• Feed the flywheel - run /retro and /post-mortem
• Prune stale knowledge - archive old artifacts

Examples

User says: /flywheel — Counts pool depths, checks recent activity, validates artifact consistency, writes health report to .agents/flywheel-status.md. Hook trigger: After /post-mortem — Compares current vs historical metrics, flags velocity drops and friction points.

Troubleshooting

Problem	Cause	Solution
All pool counts zero	.agents/ directory missing or empty	Run /post-mortem or /retro to seed knowledge pools
Velocity always zero	No recent extractions (last 7 days)	Run /retro or /post-mortem to extract and index learnings
"ao CLI not available"	ao command not installed or not in PATH	Install ao CLI or use manual pool counting fallback
Stale artifacts >50%	Long time since last session or inactive repo	Run /provenance --stale to audit and archive old artifacts

Reference Documents

• references/artifact-consistency.md
• references/promotion-tiers.md