bug-hunter

分类: 测试与安全 | 上传者: codexstar69 | 下载: 0 | 版本: v1.0（最新）
Adversarial bug hunting with a sequential-first pipeline (Recon, Hunter, Skeptic, Referee) that can optionally use safe read-only parallel triage. Finds, verifies, and auto-fixes real bugs by default (with --scan-only opt-out) using checkpointed verification and resume state for large codebases. Use this skill whenever the user wants bug finding, security audits, regression checks, or code review focused on runtime behavior.
更新日志: Source: GitHub https://github.com/codexstar69/bug-hunter
目录结构

当前层级: 根目录
📁 .claude/
- 📄 settings.local.json 58 B
📁 assets/
- 📄 pipeline-diagram.png 4.1 MB
📁 evals/
- 📄 evals.json 13.0 KB
📁 modes/
- 📄 extended.md 3.2 KB
- 📄 fix-loop.md 3.7 KB
- 📄 fix-pipeline.md 5.9 KB
- 📄 loop.md 4.8 KB
- 📄 parallel.md 2.6 KB
- 📄 scaled.md 3.0 KB
- 📄 single-file.md 1013 B
- 📄 small.md 1.8 KB
📁 prompts/
- 📄 doc-lookup.md 1.7 KB
- 📄 fixer.md 4.8 KB
- 📄 hunter.md 11.4 KB
- 📄 recon.md 6.3 KB
- 📄 referee.md 8.9 KB
- 📄 skeptic.md 7.3 KB
📁 scripts/
- 📁 tests/
  - 📁 fixtures/
    
    📄 flaky-worker.cjs 1.6 KB
    
    📄 low-confidence-worker.cjs 2.0 KB
    
    📄 success-worker.cjs 883 B
  - 📄 bug-hunter-state.test.cjs 3.6 KB
  - 📄 code-index.test.cjs 2.2 KB
  - 📄 delta-mode.test.cjs 2.2 KB
  - 📄 fix-lock.test.cjs 1.2 KB
  - 📄 payload-guard.test.cjs 1.3 KB
  - 📄 run-bug-hunter.test.cjs 7.7 KB
  - 📄 test-utils.cjs 1.4 KB
- 📄 bug-hunter-state.cjs 15.1 KB
- 📄 code-index.cjs 15.4 KB
- 📄 context7-api.cjs 3.2 KB
- 📄 delta-mode.cjs 6.0 KB
- 📄 fix-lock.cjs 3.5 KB
- 📄 init-test-fixture.sh 551 B
- 📄 payload-guard.cjs 2.7 KB
- 📄 run-bug-hunter.cjs 24.3 KB
📁 test-fixture/
- 📄 auth.js 1.7 KB
- 📄 db.js 367 B
- 📄 EXPECTED_BUGS.md 1.8 KB
- 📄 server.js 539 B
- 📄 users.js 2.0 KB
📄 .gitignore 302 B
📄 CHANGELOG.md 5.4 KB
📄 LICENSE 1.0 KB
📄 README.md 6.4 KB
📄 SKILL.md 19.1 KB
SKILL.md

---
name: bug-hunter
description: "Adversarial bug hunting with a sequential-first pipeline (Recon, Hunter, Skeptic, Referee) that can optionally use safe read-only parallel triage. Finds, verifies, and auto-fixes real bugs by default (with --scan-only opt-out) using checkpointed verification and resume state for large codebases. Use this skill whenever the user wants bug finding, security audits, regression checks, or code review focused on runtime behavior."
argument-hint: "[path | -b <branch> [--base <base-branch>] | --staged | --scan-only | --fix | --autonomous | --loop | --approve]"
disable-model-invocation: true
---

# Bug Hunt - Adversarial Bug Finding

Run a sequential-first adversarial bug hunt on your codebase. Use parallelism only for read-only triage and independent verification tasks.

## Table of Contents
- [Usage](#usage)
- [Target](#target)
- [Context Budget](#context-budget)
- [Execution Steps](#execution-steps)
- [Step 7: Present the Final Report](#step-7-present-the-final-report)
- [Self-Test Mode](#self-test-mode)
- [Error handling](#error-handling)

**Phase 1 — Find & Verify:**
```
Recon (map) --> Hunter (deep scan) --> Skeptic (challenge) --> Referee (final verdict)
                    ^                 (optional read-only dual-lens triage can run here)
                    |
             state + chunk checkpoints
```

**Phase 2 — Fix & Verify (default when bugs are confirmed):**
```
Baseline --> Git branch --> sequential Fixer (single writer) --> targeted verify --> full verify --> report
                    ^                                                              |
                    +------------------------ checkpoint commits + auto-revert -----+
```

For small scans (1-10 source files): runs single Hunter + single Skeptic (no parallelism overhead).
For large scans: process chunks sequentially with persistent state to avoid compaction drift.

## Usage

```
/bug-hunter                              # Scan entire project
/bug-hunter src/                         # Scan specific directory
/bug-hunter lib/auth.ts                  # Scan specific file
/bug-hunter -b feature-xyz              # Scan files changed in feature-xyz vs main
/bug-hunter -b feature-xyz --base dev   # Scan files changed in feature-xyz vs dev
/bug-hunter --staged                    # Scan staged files (pre-commit check)
/bug-hunter --scan-only src/            # Scan only, no code changes
/bug-hunter --fix src/                   # Find bugs AND auto-fix them
/bug-hunter --autonomous src/            # Alias for no-intervention auto-fix run
/bug-hunter --fix -b feature-xyz        # Find + fix on branch diff
/bug-hunter --fix --approve src/        # Find + fix, but ask before each fix
/bug-hunter --loop src/                  # Ralph-loop mode: audit until 100% coverage
/bug-hunter --loop --fix src/            # Loop mode: find + fix until clean
```

## Target

The raw arguments are: $ARGUMENTS

**Parse the arguments as follows:**

0. If arguments contain `--loop`: strip it from the arguments and set `LOOP_MODE=true`. The remaining arguments are parsed normally below.

0b. Default `FIX_MODE=true`.
0c. If arguments contain `--scan-only`: strip it from the arguments and set `FIX_MODE=false`.
0d. If arguments contain `--fix`: strip it from the arguments and set `FIX_MODE=true`. The remaining arguments are parsed normally below.
0e. If arguments contain `--autonomous`: strip it from the arguments, set `AUTONOMOUS_MODE=true`, and force `FIX_MODE=true` (canary-first + confidence-gated).
0f. If arguments contain `--approve`: strip it from the arguments and set `APPROVE_MODE=true`. When this flag is set, Fixer agents run in `mode: "default"` (user reviews and approves each edit). When not set, `APPROVE_MODE=false` and Fixers run autonomously.

1. If arguments contain `--staged`: this is **staged file mode**.
   - Run `git diff --cached --name-only` using the Bash tool to get the list of staged files.
   - If the command fails, report the error to the user and stop.
   - If no files are staged, tell the user there are no staged changes to scan and stop.
   - The scan target is the list of staged files (scan their full contents, not just the diff).

2. If arguments contain `-b <branch>`: this is **branch diff mode**.
   - Extract the branch name after `-b`.
   - If `--base <base-branch>` is also present, use that as the base branch. Otherwise default to `main`.
   - Run `git diff --name-only <base>...<branch>` using the Bash tool to get the list of changed files.
   - If the command fails (e.g. branch not found), report the error to the user and stop.
   - If no files changed, tell the user there are no changes to scan and stop.
   - The scan target is the list of changed files (scan their full contents, not just the diff).

3. If arguments do NOT contain `-b` or `--staged`: treat the entire argument string as a **path target** (file or directory). If empty, scan the current working directory.

**After resolving the file list (for modes 1 and 2), filter out non-source files:**

Remove any files matching these patterns — they are not scannable source code:
- `*.md`, `*.txt`, `*.rst`, `*.adoc`
- `*.json`, `*.yaml`, `*.yml`, `*.toml`, `*.ini`, `*.cfg`
- `*.lock`, `*.sum`
- `*.min.js`, `*.min.css`, `*.map`
- `*.svg`, `*.png`, `*.jpg`, `*.gif`, `*.ico`, `*.woff*`, `*.ttf`, `*.eot`
- `.env*`, `.gitignore`, `.editorconfig`, `.prettierrc*`, `.eslintrc*`, `tsconfig.json`
- `jest.config.*`, `vitest.config.*`, `webpack.config.*`, `vite.config.*`, `next.config.*`, `tailwind.config.*`
- `LICENSE`, `CHANGELOG*`, `CONTRIBUTING*`, `CODE_OF_CONDUCT*`
- `Makefile`, `Dockerfile`, `docker-compose*`, `Procfile`
- Files inside `node_modules/`, `vendor/`, `dist/`, `build/`, `.next/`, `__pycache__/`, `.venv/`

If after filtering there are zero source files left, tell the user: "No scannable source files found — only config/docs/assets were changed." and stop.

## Context Budget

Each subagent has a limited context window. The number of files an agent can reliably process depends on file sizes.

**After Recon completes, read its Context Budget section to get the dynamic FILE_BUDGET.** Recon computes this as:
```
avg_tokens_per_file = average_lines_per_file * 4
FILE_BUDGET = floor(150000 / avg_tokens_per_file)   # capped at 60, floored at 10
```

Then determine partitioning:

| Total source files | Strategy | Hunters | Skeptics |
|--------------------|----------|---------|----------|
| 1 | Single-file mode | 1 general | 1 |
| 2-10 | Small mode | 1 general | 1 |
| 11 to FILE_BUDGET | Parallel mode (hybrid) | 1 deep Hunter (+ optional 2 read-only triage Hunters) | 1-2 by directory |
| FILE_BUDGET+1 to FILE_BUDGET*2 | Extended mode | Sequential chunked Hunters | 1-2 by directory |
| FILE_BUDGET*2+1 to FILE_BUDGET*3 | Scaled mode | Sequential chunked Hunters with resume state | 1-2 by directory |
| > FILE_BUDGET*3 | Loop mode required | Sequential chunk pipeline across iterations | 1-2 by directory |

If Recon did not produce a FILE_BUDGET (e.g., Recon was skipped), use the default of 40.

**File partitioning rules (Extended/Scaled modes):**
- **Service-aware partitioning (preferred)**: If Recon detected multiple service boundaries (monorepo), partition by service.
- **Risk-tier partitioning (fallback)**: process CRITICAL then HIGH then MEDIUM.
- Keep chunk size small (recommended 20-40 files) to avoid context compaction issues.
- Persist chunk progress in `.claude/bug-hunter-state.json` so restarts do not re-scan done chunks.
- Test files (CONTEXT-ONLY) are included only when needed for intent.

If the codebase exceeds FILE_BUDGET * 3 and `--loop` was not specified, warn the user: "This codebase has [N] source files (FILE_BUDGET: [B]). For thorough coverage, consider using `--loop` mode."

## Execution Steps

### Step 0: Preflight checks

Before doing anything else, verify the environment:

1. **Resolve skill directory**: Determine `SKILL_DIR` dynamically.
   - Preferred: derive it from the absolute path of the current `SKILL.md` (`dirname` of this file).
   - Fallback probe order: `$HOME/.agents/skills/bug-hunter`, `$HOME/.claude/skills/bug-hunter`.
   - Use this path for ALL Read tool calls and shell commands.

2. **Verify skill files exist**: Run `ls "$SKILL_DIR/prompts/hunter.md"` via Bash. If this fails, stop and tell the user: "Bug Hunter skill files not found. Reinstall the skill and retry."

3. **Node.js available**: Run `node --version` via Bash. If it fails, stop and tell the user: "Node.js is required for doc verification. Please install Node.js to continue."

4. **Context7 availability (optional, non-blocking)**: Run a quick smoke test:
   ```
   node "$SKILL_DIR/scripts/context7-api.cjs" search "express" "middleware"
   ```
   - If it returns results, set `DOC_LOOKUP_AVAILABLE=true`.
   - If it fails, warn the user and set `DOC_LOOKUP_AVAILABLE=false`.
   - Missing `CONTEXT7_API_KEY` must NOT block execution; anonymous lookups may still work.

5. **Verify helper scripts exist**:
   ```
   ls "$SKILL_DIR/scripts/run-bug-hunter.cjs" "$SKILL_DIR/scripts/bug-hunter-state.cjs" "$SKILL_DIR/scripts/code-index.cjs" "$SKILL_DIR/scripts/delta-mode.cjs" "$SKILL_DIR/scripts/payload-guard.cjs" "$SKILL_DIR/scripts/fix-lock.cjs"
   ```
   If missing, stop and tell the user to update/reinstall the skill.

6. **Select orchestration backend (cross-CLI portability)**:
   - Detect available orchestration primitives in the current CLI/runtime.
   - Set `AGENT_BACKEND` using this priority:
     1. `spawn_agent` + `wait`
     2. native `subagent` delegation
     3. team-agent orchestration tools
     4. `local-sequential` (no subagents; run all phases in the main agent)
   - Use exactly ONE backend for the whole run (no mixing unless fallback is triggered).
   - If launch fails once on the selected backend, fall back to the next backend and continue.
   - If all remote backends fail, continue with `local-sequential` mode.

### Step 1: Parse arguments and resolve target

Follow the rules in the **Target** section above. If in branch diff or staged mode, run the appropriate git command now, collect the file list, and apply the filter.

Report to the user:
- Mode (full project / directory / file / branch diff / staged)
- Number of source files to scan (after filtering)
- Number of files filtered out

### Step 2: Read prompt files on demand (context efficiency)

**MANDATORY**: You MUST read prompt files using the Read tool before passing them to subagents. Do NOT skip this or act from memory. Use the absolute SKILL_DIR path resolved in Step 0.

**Load only what you need for each phase — do NOT read all files upfront:**

| Phase | Read These Files |
|-------|-----------------|
| Recon (Step 4) | `prompts/recon.md` (skip for single-file mode) |
| Hunters (Step 5) | `prompts/hunter.md` + `prompts/doc-lookup.md` |
| Skeptics (Step 6) | `prompts/skeptic.md` + `prompts/doc-lookup.md` |
| Referee (Step 7) | `prompts/referee.md` |
| Fixers (Phase 2) | `prompts/fixer.md` + `prompts/doc-lookup.md` (only if FIX_MODE=true) |

When launching subagents, always pass `SKILL_DIR` explicitly in the task context so prompt commands like `node "$SKILL_DIR/scripts/context7-api.cjs"` resolve correctly.

Before every subagent launch, validate payload shape with:
```
node "$SKILL_DIR/scripts/payload-guard.cjs" validate "<role>" "<payload-json-path>"
```
If validation fails, do NOT launch the subagent. Fix the payload first.

Any mode step that says "launch subagent" means "dispatch an agent task using `AGENT_BACKEND`".

After reading each prompt, extract the key instructions and pass the content to subagents via their system prompts. You do not need to keep the full text in working memory.

**Context pruning for subagents:** When passing bug lists to Skeptics, Fixers, or the Referee, only include the bugs assigned to that agent — not the full merged list. For each bug, include: BUG-ID, severity, file, lines, claim, evidence, runtime trigger, cross-references. Omit: the Hunter's internal reasoning, scan coverage stats, and any "FILES SCANNED/SKIPPED" metadata. This keeps subagent prompts lean.

### Step 3: Determine execution mode

Based on the scan target size, choose the execution mode per the **Context Budget** table.

Read the corresponding mode file:
- 1 file: `SKILL_DIR/modes/single-file.md`
- 2-10 files: `SKILL_DIR/modes/small.md`
- 11 to FILE_BUDGET: `SKILL_DIR/modes/parallel.md`
- FILE_BUDGET+1 to FILE_BUDGET*2: `SKILL_DIR/modes/extended.md`
- FILE_BUDGET*2+1 to FILE_BUDGET*3: `SKILL_DIR/modes/scaled.md`
- > FILE_BUDGET*3: force `LOOP_MODE=true` and read `SKILL_DIR/modes/loop.md`

If LOOP_MODE=true, also read:
- `SKILL_DIR/modes/fix-loop.md` when FIX_MODE=true
- `SKILL_DIR/modes/loop.md` otherwise

Report the chosen mode to the user.

**Then follow the steps in the loaded mode file.** Each mode file contains the specific steps for running Hunters, Skeptics, and Referee for that mode. Execute them in order.

For `extended` and `scaled` modes, initialize state before chunk execution:
```
node "$SKILL_DIR/scripts/bug-hunter-state.cjs" init ".claude/bug-hunter-state.json" "<mode>" "<files-json-path>" 30
```
Then apply hash-based skip filtering before each chunk:
```
node "$SKILL_DIR/scripts/bug-hunter-state.cjs" hash-filter ".claude/bug-hunter-state.json" "<chunk-files-json-path>"
```

For full autonomous chunk orchestration (timeouts + retries + journal), prefer:
```
node "$SKILL_DIR/scripts/run-bug-hunter.cjs" run --skill-dir "$SKILL_DIR" --files-json "<files-json-path>" --changed-files-json "<changed-files-json-path>" --mode "<mode>" --use-index true --delta-mode true --delta-hops 2 --expand-on-low-confidence true --canary-size 3 --timeout-ms 120000 --max-retries 1 --backoff-ms 1000
```
This writes:
- Run journal: `.claude/bug-hunter-run.log`
- Persistent index: `.claude/bug-hunter-index.json`
- Chunk facts: `.claude/bug-hunter-facts.json`
- Consistency report: `.claude/bug-hunter-consistency.json`
- Canary-first fix plan: `.claude/bug-hunter-fix-plan.json`

---

## Step 7: Present the Final Report

After the mode-specific steps complete, display the final report:

### 0. Verification re-audit gate (before final counts)
Run one rejection pass before locking final findings:
- Re-audit all `Critical` findings and all findings missing strong runtime trigger evidence.
- Re-audit at least 30% sample of remaining Medium/Low findings.
- Any finding that fails reproducibility or is contradicted by Skeptic/Referee in this pass must be marked `REJECTED_FALSE_POSITIVE` and excluded from confirmed totals.
- Report both counts: `raw_findings` and `confirmed_findings_after_reaudit`.

### 1. Scan metadata
- Mode (single-file / small / parallel-hybrid / extended / scaled / loop)
- Files scanned: N source files (N filtered out)
- Architecture: [summary from Recon]
- Tech stack: [framework, auth, DB from Recon]

### 2. Pipeline summary
```
Recon:     mapped N files -> CRITICAL: X | HIGH: Y | MEDIUM: Z | Tests: T | FILE_BUDGET: B
Hunters:   [deep scan findings: W | optional triage findings: T | merged: U unique]
Gap-fill:  [N files re-scanned, M additional findings] (or "not needed")
Skeptics:  [challenged X | disproved: D, accepted: A]
Referee:   confirmed N real bugs -> Critical: X | Medium: Y | Low: Z
```

### 3. Confirmed bugs table
(sorted by severity — from Referee output)

### 4. Low-confidence items
Flagged for manual review.
- Include an **Auto-fix eligibility** field per bug:
  - `ELIGIBLE`: Referee confidence >= 75%
  - `MANUAL_REVIEW`: confidence < 75% or missing confidence
- If low-confidence items exist, expand scan scope from delta mode using trust-boundary overlays before finalizing report.

### 5. Dismissed findings
In a collapsed `<details>` section (for transparency).

### 6. Agent accuracy stats
- Deep Hunter accuracy: X/Y confirmed (Z%)
- Optional triage value: N triage-only findings promoted to deep scan
- Skeptic accuracy: X/Y correct challenges (Z%)

### 7. Coverage assessment
- If ALL CRITICAL/HIGH files scanned: "Full coverage achieved."
- If any missed: list them with note about `--loop` mode.

If zero bugs were confirmed, say so clearly — a clean report is a good result.

**Routing after report:**
- If confirmed bugs > 0 AND `FIX_MODE=true`:
  - Auto-fix only `ELIGIBLE` bugs.
  - Apply canary-first rollout: fix top critical eligible subset first, verify, then continue remaining eligible fixes.
  - Keep `MANUAL_REVIEW` bugs in report only (do not auto-edit).
  - Run final global consistency pass over merged findings before applying fixes.
  - Read `SKILL_DIR/modes/fix-pipeline.md` and execute Phase 2 on eligible subset.
- If confirmed bugs > 0 AND `FIX_MODE=false`: stop after report (scan-only mode).
- If zero bugs confirmed: stop here. The report is the final output.

---

## Self-Test Mode

To validate the pipeline works end-to-end, run `/bug-hunter SKILL_DIR/test-fixture/` on the included test fixture. This directory contains a small Express app with 6 intentionally planted bugs (2 Critical, 3 Medium, 1 Low). Expected results:
- Recon should classify 3 files as CRITICAL, 1 as HIGH
- Hunters should find all 6 bugs (possibly more false positives)
- Skeptic should challenge at least 1 false positive
- Referee should confirm all 6 planted bugs

If the pipeline finds fewer than 5 of the 6 planted bugs, the prompts need tuning. If it reports more than 3 false positives that survive to the Referee, the Skeptic prompt needs tightening.

The test fixture source files ship with the skill. If using `--fix` mode on the fixture, initialize its git repo first: `bash SKILL_DIR/scripts/init-test-fixture.sh`

---

## Error handling

| Step | Failure | Fallback |
|------|---------|----------|
| Recon | timeout/error | Skip Recon, Hunters use Glob-based discovery |
| Optional triage Hunter | timeout/error | Disable triage, continue with deep Hunter |
| Deep Hunter | timeout/error | Retry once on narrowed chunk, otherwise report partial coverage |
| Orchestration backend | launch failure | Fall back to next backend (`spawn_agent -> subagent -> team -> local-sequential`) |
| Gap-fill Hunter | timeout/error | Note missed files, continue |
| Payload guard | validation fails | Do not launch subagent; fix payload and retry |
| Orchestrator phase timeout | timeout/error | Retry with exponential backoff (`max-retries`), then mark chunk failed |
| Skeptic-A | timeout/error | Run single Skeptic on all bugs |
| Skeptic-B | timeout/error | Use Skeptic-A's results, mark rest "unverified" |
| Referee | timeout/error | Use Skeptic's accepted list as final result |
| Git safety (Step 8a) | not a git repo | Warn user, skip branching |
| Git safety (Step 8a) | stash/branch fails | Warn, continue without safety net |
| Fix lock acquire | lock held | Stop Phase 2, report concurrent fixer run |
| Test baseline (Step 8c) | timeout >5min | Set BASELINE=null, cannot attribute failures |
| Test baseline (Step 8c) | command not found | Set TEST_COMMAND=null, skip test verification |
| Any single Fixer | timeout/error | Mark unfixed bugs as SKIPPED |
| Post-fix tests (Step 10a) | timeout >5min | Report "manual verification needed" |
| Post-fix tests (Step 10a) | new failures | Auto-revert failed fix commit, mark FIX_REVERTED |
| Auto-revert (Step 10b) | revert conflicts | Mark as FIX_FAILED (can't cleanly undo) |
| Post-fix re-scan (Step 10c) | timeout/error | Skip re-scan, note "fixer output not re-verified" |
| Fix lock release | release fails | Warn user to clear `.claude/bug-hunter-fix.lock` manually |
登录后下载/点赞/收藏 ❤ 10 | ★ 0
bug-hunter

目录结构

SKILL.md

举报内容

提示
