If you believe you have found a security vulnerability in this project, please report it privately by email rather than opening a public GitHub issue.

• Email: geeklin36@gmail.com
• Preferred subject line format: [SECURITY] <short description> (for example, [SECURITY] install.sh privilege escalation)

Please include a clear description of the issue, the affected component, steps to reproduce, and any proof-of-concept material you have. If the report is sensitive, you may request a PGP key in your first message.

Only the 0.1.x release line is currently supported with security fixes. Older or forked versions will not receive patches; please upgrade to the latest 0.1.x release before reporting.

The following components are in scope for security reports:

• install.sh — the bash execution surface. Anything that allows unintended command execution, path traversal, or tampering with files outside the intended install target during installation.
• Hook scripts — these have access to the Claude Code transcript. Reports involving unintended transcript exfiltration, disclosure of transcript contents, or hook-driven command execution based on attacker-controlled transcript content are in scope.
• Agent bodies (system prompts and instructions) — the LLM prompt-injection surface. Reports involving injection paths that cause the shipped agents to leak data, bypass declared restrictions, or take unsafe actions on a user's machine are in scope.

The following are not covered by this policy:

• Upstream Claude Code — please report those to Anthropic directly.
• Upstream MCP servers — please report those to the maintainers of the individual server.
• Third-party skills that users add to their own .claude/ directory. Skills authored outside this repository are the responsibility of their authors.

Denial-of-service against a user's own local machine via resource exhaustion in a locally-invoked script is generally out of scope unless it enables further compromise.

Reports are handled on a best-effort basis by a small maintainer team. We aim to acknowledge new reports within 72 hours of receipt. Triage, fix development, and release timing depend on severity and complexity, and will be communicated in the acknowledgement thread.

Disclosure is coordinated with the reporter. Once a fix is available, we will agree on a disclosure date and, where appropriate, credit the reporter in the release notes. Please do not publicly disclose the issue before a fix has been released and the coordinated date has been reached.