ResumeSAFE

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: resume
description: "Resume a previous hunt. Shows hunt history, untested endpoints, memory-informed suggestions. Usage: /resume target.com"
disable-model-invocation: false
---
Resume hunt on: $ARGUMENTS

## What This Does
1. Read brain for target: `uv run python3 $CLAUDE_PROJECT_DIR/tools/brain.py brief $ARGUMENTS`
2. Read brain exhausted: `uv run python3 $CLAUDE_PROJECT_DIR/tools/brain.py exhausted $ARGUMENTS`
3. **Read chain-pending advisory**: `cat .claude/agent-memory-local/chain-pending.md 2>/dev/null` — feeder findings from prior session waiting for chain dispatch. SURFACE THESE FIRST — they have proven dollar value if chained, vs untested surface which is speculative.
4. **Generate / refresh chain plan**: `uv run python3 $CLAUDE_PROJECT_DIR/tools/chain_plan.py $ARGUMENTS` — writes `evidence/<target>/CHAIN_PLAN.md` with 3-5 candidate next-links per pending feeder finding plus the agent to dispatch for each. Read this BEFORE picking the next vuln class — proven feeders trump speculative untested surface.
5. Read recon data from recon/ directory
6. Show what's been tested vs what remains
7. Suggest next techniques based on brain patterns

## Output
```
RESUME: target.com
═══════════════════

Brain Status:
  Confirmed findings: N
  Exhausted techniques: N
  Effective techniques: N

Untested Surface:
  1. /api/v2/users/{id}/export — not tested
  2. /api/v2/users/{id}/share — not tested
  ...

Suggestions:
  - Tech stack [X] matches pattern where IDOR was found before
  - Sibling endpoints of confirmed finding not yet tested

Actions:
  [h] /hunt target.com — resume hunting
  [s] /surface target.com — re-rank attack surface
  [r] /recon target.com — re-run recon (surface may have changed)
  [m] /monitor check — check for target changes since last session
```

## Top-Tier Resume Logic

Resume by value, not chronology.

Priority order:
1. `chain-pending` confirmed feeders with clear next link
2. validated partials needing one proof artifact
3. changed assets from `/monitor`
4. untested P1 crown-jewel endpoints
5. high-confidence class hypotheses from prior wins
6. stale recon refresh

Show what not to redo. If an area is exhausted, include the exact blocker and matrix coverage. A resumed hunt should start with a single best command, not a menu of every possible command.

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.