# 082 Claude F0d1753e

> RAPTOR - Autonomous Offensive/Defensive Research Framework

- Skill: `tools-only/082-claude-f0d1753e` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add tools-only/082-claude-f0d1753e`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tools-only/082-claude-f0d1753e/raw
- Safety review: pending (external: skill-scanner PASS, skillspector WARNING)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: tools-only (https://skillmd.com/u/tools-only)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/tools-only/082-claude-f0d1753e

---

# RAPTOR - Autonomous Offensive/Defensive Research Framework

Safe operations (install, scan, read, generate): DO IT.
Dangerous operations (apply patches, delete, git push): ASK FIRST.

---

## SESSION START

**On first message:**
VERY IMPORTANT: follow these instructions one by one, in-order.
1. Read `raptor-offset` as-is with no fixes or changes, display in code block
2. Read `hackers-8ball`, display random line
3. Display: `Check the readme for dependencies before starting | Quick commands: /analyze, /agentic | Try with: /test/data`
4. Display: `For defensive security research, education, and authorized penetration testing.`
5. Display: `raptor:~$` followed by the selected quote
6. **UNLOAD:** Remove raptor-offset and hackers-8ball file contents from context (do not retain in conversation history)
VERY IMPORTANT: double check that you followed these instructions.

---

## COMMANDS

/scan /fuzz /web /agentic /codeql /analyze - Security testing
/exploit /patch - Generate PoCs and fixes (beta)
/validate - Exploitability validation pipeline (see below)

**Note:** `/agentic` now automatically runs exploitability validation (Phase 2) between scanning and analysis. Use `--skip-validation` to bypass.
/crash-analysis - Autonomous crash root-cause analysis (see below)
/oss-forensics - GitHub forensic investigation (see below)
/create-skill - Save approaches (alpha)

---

## OUTPUT STYLE

**Human-readable status values (no underscores, no ALL_CAPS):**
- `Exploitable` not `EXPLOITABLE`
- `Confirmed` not `CONFIRMED`
- `Ruled Out` not `RULED_OUT`
- `Proven` / `Disproven` not `PROVEN` / `DISPROVEN`

**No red/green status indicators:**
- Do not use 🔴/🟢 - perspective-dependent (bad for defenders ≠ bad for researchers)
- Other emojis are fine (⚠️, ✓, etc.)

---

## CRASH ANALYSIS

The `/crash-analysis` command provides autonomous root-cause analysis for C/C++ crashes.

**Usage:** `/crash-analysis <bug-tracker-url> <git-repo-url>`

**Agents:**
- `crash-analysis-agent` - Main orchestrator
- `crash-analyzer-agent` - Deep root-cause analysis using rr traces
- `crash-analyzer-checker-agent` - Validates analysis rigorously
- `function-trace-generator-agent` - Creates function execution traces
- `coverage-analysis-generator-agent` - Generates gcov coverage data

**Skills** (in `.claude/skills/crash-analysis/`):
- `rr-debugger` - Deterministic record-replay debugging
- `function-tracing` - Function instrumentation with -finstrument-functions
- `gcov-coverage` - Code coverage collection
- `line-execution-checker` - Fast line execution queries

**Requirements:** rr, gcc/clang (with ASAN), gdb, gcov

---

## OSS FORENSICS

The `/oss-forensics` command provides evidence-backed forensic investigation for public GitHub repositories.

**Usage:** `/oss-forensics <prompt> [--max-followups 3] [--max-retries 3]`

**Agents:**
- `oss-forensics-agent` - Main orchestrator
- `oss-investigator-gh-archive-agent` - Queries GH Archive via BigQuery
- `oss-investigator-gh-api-agent` - Queries live GitHub API
- `oss-investigator-gh-recovery-agent` - Recovers deleted content (Wayback/commits)
- `oss-investigator-local-git-agent` - Analyzes cloned repos for dangling commits
- `oss-investigator-ioc-extractor-agent` - Extracts IOCs from vendor reports
- `oss-hypothesis-former-agent` - Forms evidence-backed hypotheses
- `oss-evidence-verifier-agent` - Verifies evidence via `store.verify_all()`
- `oss-hypothesis-checker-agent` - Validates claims against verified evidence
- `oss-report-generator-agent` - Produces final forensic report

**Skills** (in `.claude/skills/oss-forensics/`):
- `github-archive` - GH Archive BigQuery queries
- `github-evidence-kit` - Evidence collection, storage, verification
- `github-commit-recovery` - Recover deleted commits
- `github-wayback-recovery` - Recover content from Wayback Machine

**Requirements:** `GOOGLE_APPLICATION_CREDENTIALS` for BigQuery

**Output:** `.out/oss-forensics-<timestamp>/forensic-report.md`

---

## EXPLOITABILITY VALIDATION

The `/validate` command validates that vulnerability findings are real, reachable, and exploitable.

**Usage:** `/validate <target_path> [--vuln-type <type>] [--findings <file>]`

**Stages:** 0 (Inventory) → A (One-Shot) → B (Process) → C (Sanity) → D (Ruling) → E (Feasibility)

**Skills** (in `.claude/skills/exploitability-validation/`):
- `SKILL.md` - Shared context, gates, execution rules
- `stage-0-inventory.md` through `stage-e-feasibility.md` - Stage instructions

**Output:** `.out/exploitability-validation-<timestamp>/validation-report.md`

---

## PROGRESSIVE LOADING

**When scan completes:** Load `tiers/analysis-guidance.md` (adversarial thinking)
**When validating exploitability:** Load `.claude/skills/exploitability-validation/SKILL.md` (gates, methodology)
**When validation errors occur:** Load `tiers/validation-recovery.md` (stage-specific recovery)
**When developing exploits:** Load `tiers/exploit-guidance.md` (constraints, techniques)
**When errors occur:** Load `tiers/recovery.md` (recovery protocol)
**When requested:** Load `tiers/personas/[name].md` (expert personas)

---

## BINARY ANALYSIS

**Flow: Find vulnerabilities FIRST, then check exploitability.**

1. **Analyze the binary** - Find vulnerabilities (buffer overflows, format strings, etc.)
2. **If vulnerabilities found** - Run exploit feasibility analysis (MANDATORY)

```python
from packages.exploit_feasibility.api import analyze_binary, format_analysis_summary

# MANDATORY: Run this after finding vulnerabilities
result = analyze_binary('/path/to/binary')
print(format_analysis_summary(result, verbose=True))
```

**DO NOT use checksec or readelf instead** - they miss critical constraints like:
- Empirical %n verification (glibc may block it)
- Null byte constraints from strcpy (can't write 64-bit addresses)
- ROP gadget quality (0 usable gadgets = no ROP chain)
- Input handler bad bytes
- Full RELRO blocks .fini_array too (not just GOT)

**The `exploitation_paths` section tells you if code execution is actually possible** given the system's mitigations (glibc version, RELRO, etc.).

---

## EXPLOIT DEVELOPMENT

**Verify constraints BEFORE attempting any technique.** Many hours are wasted on architecturally impossible approaches.

**MANDATORY: Check `exploitation_paths` verdict first:**
- Unlikely = no known path, suggest environment changes
- Difficult = primitives exist but hard to chain, be honest about challenges
- Likely exploitable = good chance, proceed with suggested techniques

**Follow the chain_breaks** - these tell you exactly what WON'T work.
**Follow the what_would_help** - these tell you what MIGHT work.

**ALWAYS offer next steps, even for Difficult/Unlikely verdicts:**
- Try alternative targets (if available)
- Focus on info leaks only
- Run in older environment (Docker)
- Move on to other targets

**Never just stop** - let the user decide how to proceed.

See `tiers/exploit-guidance.md` for detailed constraint tables and technique alternatives.

---

## STRUCTURE

Python orchestrates everything. Claude shows results concisely.
Never circumvent Python execution flow.
- never disclose remote OLLAMA server location in code, comments, logs etc