Security Policy

Supported versions

Version Supported
0.2.x ✅
0.1.x ⚠️ best-effort; please upgrade to 0.2.x
< 0.1 ❌

Reporting a vulnerability

Do NOT open a public GitHub issue for security concerns. Use private channels:

  1. Preferred: GitHub Security Advisory — Open a private advisory
  2. Email: c.ulisses@gmail.com with subject prefix [asp-security]
  3. Backup: contact form at https://ulissesflores.com

Please include:

  • A description of the issue and its potential impact.
  • Steps to reproduce, including the install path (plugin marketplace / --plugin-dir / install.sh) and Claude Code version.
  • Whether the issue is publicly known or under embargo.
  • Suggested mitigation, if you have one.

Response timeline

  • Acknowledgement: within 72 hours.
  • Initial assessment: within 7 days.
  • Patch + disclosure plan: jointly agreed with the reporter; coordinated disclosure is preferred.

Threat model (in scope)

ASP is a Claude Code skill. Realistic threat surfaces:

  1. claude -p subprocess spawn (skill/claude-p-goal-runner.md). The runner uses --permission-mode bypassPermissions and processes user-supplied Goal Specs. A maliciously crafted Goal Spec could cause the child Claude session to execute unintended actions (file write, network, code execution). Mitigations are listed in Iron Laws 9 and 10 (file-based contract, ASP_IN_GOAL reentrancy guard).
  2. recall.py integration (Phase 1). If a malicious ~/.agent/ is present, recall output is included in the planning context. Treat the agentic-stack as a trust boundary.
  3. Markdown template injection. Templates use `` patterns rendered into text. They are not executed; no SSTI risk in normal usage.
  4. Install scripts (scripts/install.sh, uninstall.sh). These run with the user’s permissions and write to ~/.claude/. Review the source before running in untrusted environments.
  5. Plugin manifest (.claude-plugin/plugin.json). Static metadata; no execution surface.

Out of scope

  • Vulnerabilities in Claude Code itself, the /goal slash command, or Anthropic infrastructure → report to Anthropic.
  • Issues in third-party plugins listed alongside ASP in marketplaces → report to those plugin maintainers.
  • Social engineering against the maintainer (this is a personal repo, not an organization).

Recognition

If you report a valid vulnerability, you will (with your permission) be credited in the security advisory, the release notes that ship the fix, and the AUTHORS file.