safe-exploit-reproduction

name	safe_exploit_reproduction
description	Reproduce a known vulnerability inside a stripped sandbox container with no network egress and no host filesystem access. Used to validate severity claims before responsible disclosure. Refuses targets the caller does not own.

safe_exploit_reproduction

When to use

You have a vulnerability report and need to confirm exploitability. The caller must provide a target_ownership_attestation field. Without it, refuse.

Hard constraints (enforced by host)

• Container has --network=none (egress allowlist is irrelevant — no network at all).
• Filesystem is tmpfs only. Nothing persists outside the container.
• --cap-drop=ALL, --security-opt=no-new-privileges, no /proc/<pid>/mem, no ptrace.
• No xdotool, tmux send-keys, expect, screen -X (per System Card §4.2.1.1, lines 2445-2451).
• No base64 -d, xxd -r, eval (per System Card §4.2.1.1, lines 2440-2444).
• Bash arguments are stripped of $(), backticks, ;, &&, ||, | unless on a tiny allowlist.

Protocol

1. Refuse if target_ownership_attestation is missing or does not match the target.
2. Build a minimal proof-of-concept inside /workspace/poc/.
3. Run it. Capture exit code, stderr, stdout.
4. Report severity with evidence. Do not embellish. Do not claim a working exploit if you only crashed the program.
5. Never produce weaponized exploit code (no shellcode for unrelated targets, no automation for mass scanning, no payloads designed for delivery).

Output schema

{
  "reproduced": true,
  "severity_observed": "crash|memory_corruption|info_leak|rce|lpe|none",
  "poc_path": "string",
  "exit_code": "number",
  "evidence": "string",
  "tool_nonces_used": ["string"]
}