name: bughunt
description: Deep multi-agent bug analysis. 6 personas × N zones → findings report + inbox items for Spark. Triggers on keywords: bug hunt, deep analysis, охота на баги, баг-хант, командный аудит багов
Bug Hunt — Standalone Deep Analysis Skill
Multi-agent pipeline for systematic bug detection across codebase zones.
Activation: bughunt, bug hunt, deep analysis, охота на баги
When to Use
- Complex/systemic bugs affecting >5 files
- Proactive quality sweep of a module or domain
- User explicitly requests "bug hunt" or "deep analysis"
- Post-refactoring verification
Don't use: Simple bugs (<5 files) → use /spark Quick Bug Mode instead.
Architecture
Bughunt manages ALL pipeline steps directly at Task nesting Level 1.
Reuses existing agents from .claude/agents/bug-hunt/.
Output: Report + N inbox files (NOT backlog entries — Spark creates specs).
Step 0: scope-decomposer → zones.yaml
Step 1: 6 personas × N zones (background fan-out) → findings/
Step 2: findings-collector → summary.yaml
Step 3: spec-assembler → umbrella report
Step 4: validator → grouped findings
Step 5: report-updater → final report
Step 6: Create N inbox files (one per finding group)
Modules
| Module | Content |
|---|
pipeline.md | Full 7-step pipeline with file gates and ADR compliance |
completion.md | Inbox output format, commit + push, cleanup |
Flow: SKILL.md → pipeline.md → completion.md
STRICT RULES
- READ-ONLY for source code — Bughunt NEVER modifies application files
- No backlog entries — Bughunt writes to inbox, Spark creates specs
- No autopilot handoff — Bughunt finishes after inbox + push
- All agents background — ADR-009 compliance
- File gates — Each step verifies previous step's output before proceeding
Headless Mode
When [headless] marker present in prompt:
- Skip cost confirmation (already approved via Telegram)
- Proceed directly with pipeline
Cost Estimate
~$15/zone (6 Sonnet personas) + $10 fixed (validator + architects) = **$30-50 typical**
Print estimate before launch (non-blocking — don't wait for confirmation).
