/bug-scan - Commit Bug Scanner

Proactively scan recent commits for likely bugs, logic errors, security issues, and regressions. Analyzes diffs, identifies suspicious patterns, and proposes minimal fixes with evidence.

System Prompt

EXECUTION DIRECTIVE: When the user invokes this command, you MUST:

1. IMMEDIATELY execute - no questions first
2. Load the agent commit-bug-scanner
3. Determine scope from args or default to last 24h
4. Collect diffs for all commits in range
5. Apply pattern detection (null access, off-by-one, resource leaks, etc.)
6. Classify findings by severity and confidence
7. Propose fixes for each finding
8. Output structured scan report

---

Usage

# Default: last 24 hours
/bug-scan

# Custom time window
/bug-scan --since "3 days ago"

# Specific SHA range
/bug-scan --sha-range abc123..def456

# Scope to directory
/bug-scan --path src/api/

# Higher confidence threshold
/bug-scan --min-confidence 0.8

# Filter by severity
/bug-scan --severity critical,high

Options

Option	Description	Default
--since	Lookback window	24h
--sha-range	Explicit SHA range	none
--path	Scope to directory	.
--min-confidence	Minimum confidence to report	0.6
--severity	Filter by severity level	all
--ignore-pattern	Skip file patterns	none

Related Commands

• /triage - Triage and classify detected bugs
• /regression-check - Check for test regressions
• /security-scan - Deep security-focused analysis
• /test-generate - Generate tests for detected patterns

Success Output

COMMAND COMPLETE: /bug-scan
Commits Scanned: 15 (last 24h)
Findings: 4 (1 critical, 2 high, 1 medium)
Top Finding: Null access in src/api/handler.py:45 (confidence: 0.92)
Fixes Proposed: 4

Completion Checklist

• Commit range determined
• Diffs collected
• Bug patterns applied
• Findings classified by severity
• Confidence scores assigned
• Fix proposals generated

Failure Indicators

• No commits in specified range
• Unable to parse diffs
• All findings below confidence threshold

When NOT to Use

• Full security audit (use /security-scan)
• Code review (use /local-review or /full-review)
• Style/linting issues (use linter directly)

Anti-Patterns

Anti-Pattern	Problem	Solution
Scan without context	High false positives	Consider surrounding code
Report style issues	Signal dilution	Only functional bugs
Low confidence threshold	Alert fatigue	Keep >= 0.6

Principles

• #3 Complete Execution - Full scan from collection to fix proposals
• #9 Based on Facts - Every finding backed by diff evidence

Full Standard: CODITECT-STANDARD-AUTOMATION.md