ICR Audit

Perform deep analysis of receipt history to identify patterns, risks, and recommendations.

Usage

• /icr:audit - Audit current session
• /icr:audit session - Same as above
• /icr:audit 2026-01-01:2026-01-02 - Audit date range
• /icr:audit all - Audit all receipts

Analysis Includes

Decision Distribution

How actions were routed:

• Auto-approved (went through without review)
• AI reviewed (AI reviewer evaluated)
• Human reviewed (user made decision)
• Blocked (user or AI rejected)

Severity Distribution

Breakdown of action risk levels:

• LOW (trivial, easily reversible)
• MEDIUM (moderate impact)
• HIGH (significant impact)
• CRITICAL (dangerous, hard to reverse)

Confidence Statistics

• Average confidence score
• Min/max confidence (with receipt IDs)
• Distribution across ranges

Patterns of Interest

• Bypasses logged (user overrode recommendations)
• Blocks after AI flagged concerns
• Low confidence patterns
• Historical pattern match rates

Recommendations

Based on the patterns found:

• Suggested user rules for common patterns
• Trust mode usage suggestions
• Threshold tuning recommendations

Example Output

ICR Audit Report
---------------------------------------------------------------------
Period: 2026-01-02 (1 session, 47 receipts)

DECISION DISTRIBUTION:
  Auto-approved:  32 (68%)  ####################..........
  AI reviewed:     8 (17%)  #####.........................
  Human reviewed:  5 (11%)  ###...........................
  Blocked:         2 (4%)   #.............................

SEVERITY DISTRIBUTION:
  LOW:      25 (53%)  ################..............
  MEDIUM:   15 (32%)  ##########....................
  HIGH:      5 (11%)  ###...........................
  CRITICAL:  2 (4%)   #.............................

CONFIDENCE ANALYSIS:
  Average:  0.74
  Min:      0.42 (receipt #023, Bash "rm -rf ...")
  Max:      0.95 (receipt #001, Write "config.json")

PATTERNS OF INTEREST:
  3 bypasses logged (receipts #012, #034, #041)
  2 blocks after AI review flagged concerns
  Historical pattern match rate: 78%

RECOMMENDATIONS:
  - Consider adding user rule for "rm" commands (2 CRITICAL reviews)
  - Trust mode was used 0% of time; consider enabling for routine work
  - Average user response time: 2.3s (healthy)
---------------------------------------------------------------------