skillet is a roster of agent skills (plain SKILL.md files), not a running service. Only the latest tagged release on the main branch receives fixes. Pin to a released tag if you need a known-good version.

Please do not open a public GitHub issue for security problems. Email me@solomonneas.dev with:

• A short description of the issue.
• Steps to reproduce (or a minimal proof of concept).
• The skill and the tag or commit you tested against.
• Whether you would like to be credited in the release notes.

You should get an acknowledgment within 72 hours. If you do not, please follow up; the mail may have been filtered.

• A skill that instructs an agent to leak secrets, exfiltrate data, or run destructive commands without the documented human-in-the-loop gate.
• A skill whose procedure causes an agent to write outside the repo it was pointed at, or to push, publish, or release without explicit operator approval.
• Prompt-injection or untrusted-content handling flaws in a skill that processes external input (for example the leak-scrub and audit skills). In-scope skills must declare the shared stance in docs/untrusted-content.md.
• A skill that bypasses its own stated read-only or review-gated boundary (for example the audit trio applying changes, or release-cut tagging without a request).

A skill is a set of instructions an AI coding agent follows. The agent runs commands on your machine with your permissions. Treat any skill as only as trustworthy as the harness and the operator approvals around it:

• The read-only audit skills (line-check, bug-hunt, security-sweep, special) report findings; expedite is the separate, opt-in step that applies fixes, and it parks anything destructive or breaking for you to decide.
• Publish and release skills (publish-readiness, release-cut, pass) gate on operator review and never push or tag without an explicit request.
• The real boundary is your harness's permission model and your review of what the agent is about to do. Read the skill before you let an agent act on it.

• Bugs in content-guard itself; please report those upstream at https://github.com/escoffier-labs/content-guard.
• Bugs in Claude Code, Codex, OpenClaw, or any other harness; report those to their respective projects.
• Issues that require an attacker to already have write access to your machine, harness config, or repository.
• Content a user wrote and committed themselves. skillet provides procedures and guardrails, not perfect content review.

We aim to ship a fix within 14 days of confirming a valid report. A coordinated disclosure timeline can be negotiated for issues that need longer.