agent-docs is on 0.x. Only the latest minor version receives fixes.

Please do not open a public issue for security-relevant problems.

Email k.kreuter@it-brb.de with:

• a short description of the issue,
• the affected file(s) or command(s),
• a minimal reproduction if possible.

You'll get an acknowledgement within 72 hours and a status update within 7 days.

agent-docs is a Claude Code skill. It:

• reads files in the repo where the user invokes it,
• writes Markdown files to .agent-docs/ in that repo,
• shells out to git and python3,
• spawns Claude Code sub-agents (Explore type).

It does not:

• make outbound network requests on its own,
• read files outside the repo (beyond the skill's own directory under ~/.claude/skills/agent-docs/),
• modify git state (no commits, no pushes, no branch ops),
• collect or transmit telemetry.

Areas worth scrutinizing in a report:

• Argument-injection paths into git or python3 (e.g. via crafted topic slugs).
• Path-traversal in the topic-to-filename mapping.
• Unintended behavior when .agent-docs/ contains symlinks or non-Markdown files.
• Anything that causes the skill to write outside .agent-docs/.