tk-github-image-upload-to-pr

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads and installs the gh-attach extension for the GitHub CLI. It references the author's fork MTGVim/gh-attach and an upstream version enthus-appdev/gh-attach. The skill implements a trust preflight that inspects gh extension list to classify these sources and requires user consent for versions not explicitly reviewed by the author.
  • [REMOTE_CODE_EXECUTION]: The skill executes remote code by installing and running GitHub CLI extensions. This is managed through strict version pinning (--pin v0.7.0-mtgvim.1) and user-level authorization for any unreviewed distributions.
  • [COMMAND_EXECUTION]: The skill executes various system commands including gh extension list, gh attach, gh auth status, git check-ignore, and git ls-files. It also uses browser automation through Chrome DevTools Protocol (CDP) and Playwright for the fallback route.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub PR bodies and comments, creating a potential surface for indirect injection.
  • Ingestion points: Reads existing PR bodies and comment content via GitHub API or browser (SKILL.md, references/gh-attach.md).
  • Boundary markers: Includes a mandatory "Upload mutation boundary" checkpoint to verify targets, authentication, and requested write scope before modifying state.
  • Capability inventory: Can update PR bodies/comments, execute CLI extensions, and perform network operations via browser automation.
  • Sanitization: Verifies rendered HTML/page evidence after updates and redacts signed parameters or JWTs from asset URLs.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 05:55 AM
Security Audit — agent-trust-hub — tk-github-image-upload-to-pr