/audit

/audit — Brain Health Review

Instruction

Audit the current brain's memory quality: $ARGUMENTS

If no specific focus given, run full audit across all 6 dimensions.

Method

Step 1: Collect Baseline

nmem_stats          → neuron count, synapse count, memory types, age distribution
nmem_health         → purity score, component scores, warnings, recommendations
nmem_context        → recent memories, freshness indicators
nmem_conflicts(action="list") → active contradictions

Step 2: Score 6 Dimensions

Purity (25%)

• Active contradictions (CRITICAL if >0)
• Near-duplicates (HIGH)
• Outdated facts (MEDIUM)

Freshness (20%)

• Stale ratio (% memories >90 days with no access)
• Expired TODOs still active
• Zombie memories (never recalled, >30 days)

Coverage (20%)

• Topic balance across project areas
• Decision coverage (reasoning stored?)
• Error pattern documentation

Clarity (15%)

• Vague memories (<20 chars, no specifics)
• Missing context (decisions without reasoning)
• Overstuffed memories (>500 chars, 3+ topics)

Relevance (10%)

• Orphaned project references
• Technology drift
• Context mismatch

Structure (10%)

• Connectivity (orphan neurons)
• Synapse diversity
• Fiber conductivity

Step 3: Triage Findings

Severity	Criteria
CRITICAL	Active contradictions, security-sensitive errors
HIGH	Significant gaps, widespread staleness
MEDIUM	Moderate quality issues, some duplicates
LOW	Cosmetic, minor optimization
INFO	Observations, no action needed

Step 4: Generate Report

Memory Audit Report
Brain: {name} | Date: {today}

Overall Grade: {letter} ({score}/100)

Dimension Scores:
  Purity:     {bar}  {score}/100
  Freshness:  {bar}  {score}/100
  Coverage:   {bar}  {score}/100
  Clarity:    {bar}  {score}/100
  Relevance:  {bar}  {score}/100
  Structure:  {bar}  {score}/100

Findings: {total}
  CRITICAL: {n}
  HIGH:     {n}
  MEDIUM:   {n}
  LOW:      {n}

Top Recommendations:
  1. [{severity}] {recommendation}
  2. [{severity}] {recommendation}
  3. [{severity}] {recommendation}

Rules

• Evidence-based only — every finding must reference specific memories or metrics
• No guessing — if data is insufficient, report "insufficient data" for that dimension
• Prioritize by impact — present CRITICAL before LOW
• Actionable recommendations — every finding must have a concrete fix
• Audit is read-only — user decides what to fix
• Estimate effort for each recommendation