diagnose

/diagnose — Gate Effectiveness Analysis

When to use

When user says "diagnose", "gate health", "which gates fire most", "gate effectiveness", "analyze gates", "gate stats", "gate report", or wants to understand how framework gates are performing.

Commands

• /diagnose — Full gate effectiveness analysis for today
• /diagnose --date YYYY-MM-DD — Analyze a specific day's audit log

---

Phase 1: COLLECT — Read Today's Audit Log

Determine today's date and read the audit log:

from datetime import date
today = date.today().isoformat()  # e.g., "2026-02-20"
audit_path = f"~/.claude/hooks/audit/{today}.jsonl"

Read the file line by line (it is newline-delimited JSON). If the file does not exist, report "No audit log found for {today}" and stop.

Also read the gate effectiveness tracker:

~/.claude/hooks/.gate_effectiveness.json

For each audit log entry, extract:

• gate — gate name/label (e.g., "GATE 1: READ BEFORE EDIT")
• decision — "block", "warn", or "pass"
• tool — which tool triggered the gate
• timestamp — for timing analysis
• session_id — to identify multi-agent context

Count per gate:

• Total fires
• Blocks (decision == "block")
• Warns (decision == "warn")
• Passes (decision == "pass")

---

Phase 2: ANALYZE — Calculate Effectiveness Metrics

For each gate compute:

1. Block rate = blocks / total_fires (if total_fires > 0)
2. Warn rate = warns / total_fires
3. False positive indicator: high blocks (>10) + no corresponding test failures in state → flag as potentially noisy
4. Status label:
   • EFFECTIVE — block_rate > 0.05 and total_fires > 5
   • NOISY — block_rate > 0.30 and total_fires > 20 (fires a lot, blocks often)
   • QUIET — total_fires < 3 (rarely triggered)
   • WARN-ONLY — warns > 0 and blocks == 0
   • PASSING — passes == total_fires (never blocked or warned)

Also load the .gate_effectiveness.json cumulative data to cross-reference lifetime block counts vs today's counts.