Skip to content

Security: t3chn/skillregistry

Security

docs/SECURITY.md

title status
Security Model
active

Security Model

Threat model

Skills and plugins influence agent behavior and may execute scripts. Treat them as a supply-chain surface.

Trusted sources (policy)

  • Bootstrap must install templates only from an allowlisted skillregistry (Git URL or local path).
  • Do not fetch skills from public catalogs as part of bootstrap.

Script-backed skills

  • Prefer instruction-only skills.
  • If scripts are required:
    • keep them minimal and deterministic,
    • avoid network access unless explicitly required and reviewed,
    • avoid executing arbitrary shell commands constructed from untrusted input.

Large artifacts

  • Never paste large logs/specs into SKILL.md.
  • Store them under references/ and link to them.

Project hygiene (recommendation)

In consuming project repos, ignore:

  • .agent/skillregistry/
  • .agent/project_profile.json
  • .agent/skills_state.json
  • .agent/skills_todo.md
  • .agent/overlays_pending/

Commit project-owned overlays if desired (project-workflow, api-*).

There aren’t any published security advisories