メニュー
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
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| 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 |
Multi-agent pipeline for systematic bug detection across codebase zones.
Activation: bughunt, bug hunt, deep analysis, охота на баги
Don't use: Simple bugs (<5 files) → use /spark Quick Bug Mode instead.
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)
| 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
When [headless] marker present in prompt:
~$15/zone (6 Sonnet personas) + $10 fixed (validator + architects) = **$30-50 typical**
Print estimate before launch (non-blocking — don't wait for confirmation).
Upgrade DLD framework files from latest template on GitHub.
Autonomous spec execution with subagents (planner, coder, tester, reviewer).
Autonomous spec execution with subagents (planner, coder, tester, reviewer).
Feature specification and research agent. Multi-agent with 4 scouts. Creates specs in ai/features/.
Feature specification and research agent. Multi-agent with 4 scouts. Creates specs in ai/features/.
Manual QA tester — tests product behavior like a real user, not code. Triggers on keywords: test, QA, check behavior, verify feature, manual testing, протестируй, потыкай, проверь как работает