CorrelateSAFE

Bug bounty agent framework for Claude Code, Codex, Gemini, Cursor, Windsurf, Copilot, and OpenClaw — 48 agents, 26 commands, 19 CLI tools, 2 MCP servers, autonomous hunt loops, exploit chain builder.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: correlate
description: "Run the finding correlation engine to discover attack chains from individual findings."
disable-model-invocation: false
---
Find attack chains by correlating all known findings.

1. `uv run python3 $CLAUDE_PROJECT_DIR/tools/statusline.py --compact` — show current state
2. Launch `correlator` agent: "Analyze ALL findings in brain and findings.json. Find attack chains where one finding enables another. Document each chain with combined CVSS 4.0 and end-to-end reproduction steps."
3. After agent returns, update brain with discovered chains.
4. Show the user any new high-impact chains found.

## Top-Tier Correlation Bar

Build a capability graph, not a list of related bugs.

- Nodes are capabilities: read tenant data, write config, trigger webhook, steal token, reach internal host, execute workflow.
- Edges require evidence that one capability enables the next. Shared component or same endpoint family is only a hint.
- Score each chain by final impact, proof reliability, policy safety, duplicate risk, and report clarity.
- Prefer chains that convert low-severity feeders into account takeover, tenant escape, stored XSS with privileged action, SSRF to credential disclosure, or config write to execution.
- Record killed chains too. The next correlation pass should know which attractive edges failed and why.

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

Gates applied: no_behavioural_pass, no_license.