TriageSAFE

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: triage
description: "Batch-validate ALL findings through the 7-Question Gate. Kills weak findings in bulk. Usage: /triage"
disable-model-invocation: false
---
Batch triage all findings.

ALL validator agents dispatched by this command MUST use `model: "inherit"`.

## Process

1. Read findings.md and/or findings.json
2. List all findings with a numbered summary
3. For EACH finding, launch `validator` agent with the finding details
4. Collect results: PASS / KILL / DOWNGRADE / CHAIN REQUIRED
5. Output summary table:

```
TRIAGE RESULTS
═══════════════
#  Finding                              Decision    Reason
1  GraphQL schema leakage               KILL Q7     Never-submit: introspection alone
2  Config exposure SayTech               KILL Q7     SPA client config is by design
3  Internal service URLs                 KILL Q6     Not exploitable externally
4  IDOR on /api/users/{id}              PASS        Confirmed with real data
5  XSS on comments                      PASS        Cookie theft PoC works

PASSED: 2 findings → ready for /report
KILLED: 3 findings → removed from queue
```

6. Update brain with triage results
7. For KILLED findings: `uv run python3 $CLAUDE_PROJECT_DIR/tools/brain.py record <target> exhausted "<finding>" "<kill reason>"`
8. For PASSED findings: suggest `/report` or `/validate` for full PoC + evidence

## Top-Tier Triage Standard

Batch triage should reduce the queue aggressively.

For each finding, produce:
- decision: PASS, KILL, DOWNGRADE, CHAIN REQUIRED, DUPCHECK REQUIRED, EVIDENCE REQUIRED
- deciding gate: the first question or artifact that controlled the outcome
- missing proof: exact command, account, request, browser check, or chain needed
- reportability: bounty-grade, pentest-note, internal hardening, or discard
- memory action: confirmed, exhausted, partial, duplicate-risk, or chain-pending

Do not average weak findings into a stronger story. Chain them only when one finding provides a capability the next finding consumes.

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.