Skip to content

Security: aihxp/godpowers

Security

SECURITY.md

Security Policy

Reporting a Vulnerability

If you discover a security vulnerability in Godpowers, please report it responsibly.

How to Report

  1. Do NOT open a public GitHub issue
  2. Use GitHub's private vulnerability reporting: https://github.com/aihxp/godpowers/security/advisories/new
  3. Include:
    • Description of the vulnerability
    • Steps to reproduce
    • Potential impact
    • Suggested remediation if you have one

What to Expect

  • Best-effort acknowledgment, typically within 7 days (this is a small, pre-launch project, so treat these as targets, not guarantees)
  • Best-effort assessment, typically within 14 days
  • Fix timeline based on severity
  • Credit in the CHANGELOG when the fix ships (unless you prefer anonymity)

Scope

Godpowers is a meta-prompting framework. Security concerns include:

In scope

  • Vulnerabilities in bin/install.js (file system access, path traversal)
  • Vulnerabilities in hooks/*.sh (command injection, privilege escalation)
  • Vulnerabilities in scripts/*.{sh,js} (CI / test infrastructure)
  • Skill or agent prompts that could be exploited to leak credentials

Out of scope

  • AI model behavior (report to the model provider)
  • Issues in dependencies (report upstream)
  • Social engineering of AI agents (use --conservative mode)

Hardening Recommendations

When using Godpowers in a sensitive context:

  1. Review --yolo decisions: Before merging or deploying, read .godpowers/YOLO-DECISIONS.md to verify auto-picked defaults match intent
  2. Never accept Critical findings under --yolo: This is enforced by the framework but worth re-checking
  3. Keep .godpowers/ out of public repos if it contains sensitive PRD content (add to .gitignore per-project)
  4. Hooks are advisory, not a sandbox: hooks/pre-tool-use.sh and hooks/session-start.sh run with your shell privileges. The pre-tool-use hook only warns on common destructive command spellings (it is a typo guard and is easily bypassed by uncommon spellings, quoting, aliases, or a child process); do not rely on it as a security boundary. Review both before installing.
  5. Verify the npm package signature: npm audit signatures (verifies registry provenance and the published package signature)
  6. Treat .godpowers/ledger/ as executable, output-bearing state: the evidence ledger records the exact commands you run via verify/outcome plus tails of their stdout/stderr. If a command or its output can contain a secret, add .godpowers/ledger/ to .gitignore so it is not committed. The outcome check command re-runs a verifier stored in goal.json, so only run it in repositories you trust.
  7. Codex agents install with sandbox_mode = "workspace-write": the Codex runtime grants every installed Godpowers agent write access to the workspace (they need it to write artifacts). Combined with untrusted instructions in project files, an agent could write anywhere in the workspace; narrow the Codex sandbox per agent if that is a concern.

Supported Versions

Version Supported
3.14.x Yes
3.13.x Security fixes only
3.12.x Security fixes only
3.11.x Security fixes only
3.10.x Security fixes only
3.9.x Security fixes only
3.8.x Security fixes only
3.7.x Security fixes only
3.6.x Security fixes only
3.5.x Security fixes only
3.4.x Security fixes only
3.3.x Security fixes only
3.2.x Security fixes only
3.1.x Security fixes only
3.0.x Security fixes only
2.7.x Security fixes only
2.6.x Security fixes only
2.5.x Security fixes only
2.4.x Security fixes only
2.3.x Security fixes only
2.2.x Security fixes only
2.1.x Security fixes only
< 2.1 No

Godpowers repo documentation sync checks this table as part of release readiness, but support policy changes still require maintainer review.

Disclosure Policy

We follow coordinated disclosure:

  1. Reporter privately reports the issue
  2. We acknowledge within 7 days
  3. We work on a fix
  4. We coordinate disclosure timing with the reporter
  5. Public disclosure happens after the fix is released

We aim for fix-to-disclosure within 90 days for most issues, faster for Critical severity.

There aren't any published security advisories