Skip to content

Security: AdrianParedez/capability-fabric

.github/SECURITY.md

Security

Trust model

Capability Fabric is data, not a service. Skills are plain Markdown and supporting files that your agent reads on demand. The library makes no network calls and collects no telemetry.

The installer runs locally and does two things: it copies the skill folders into each agent's skills directory, and it writes one marker-delimited block into each agent's global config (~/.claude/CLAUDE.md, ~/.codex/AGENTS.md, ~/.agents/AGENTS.md). It changes nothing else. -Uninstall / --uninstall reverses both and leaves other content intact.

Skills can instruct an agent to run commands. Review a skill's SKILL.md and any scripts/ before installing it, the same way you would review any code you run. Treat third-party skills with the same caution as third-party code.

Reporting a vulnerability

Report privately. Do not open a public issue for a security problem.

  • Preferred: GitHub private vulnerability reporting, the Security tab on this repository, then Report a vulnerability.
  • Include: what you found, how to reproduce it, and the impact.

You will get an acknowledgement within a few days. Fixes are released as soon as a verified report is confirmed, with credit to the reporter unless you prefer otherwise.

Scope

In scope: the installer scripts, the routing block, and any bundled scripts/. Out of scope: behaviour of the host agent runtime itself, and skills you author or install from elsewhere.

There aren't any published security advisories