The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes external GitHub CLI extensions (enthus-appdev/gh-attach and MTGVim/gh-attach). It incorporates a mandatory security preflight check where the agent must verify the specific Git commit SHA of the extension against hardcoded trusted values before performing installation or execution.
[REMOTE_CODE_EXECUTION]: The skill provides instructions for installing and running a GitHub CLI extension (gh-attach). This execution is protected by commit-pinning and a verification process that requires checking the remote repository state via API before use.
[COMMAND_EXECUTION]: The skill invokes the gh CLI for repository interactions and manages local file operations (staging and cleanup) using absolute paths to prevent accidental deletions or command injection via malformed filenames.
[INDIRECT_PROMPT_INJECTION]: The skill reads and processes existing GitHub PR bodies and comments, which may contain untrusted data.
Ingestion points: Workflow steps in SKILL.md involve reading the existing PR body to identify insertion points.
Boundary markers: The skill does not use unique delimiters but follows a fixed workflow to "Preserve unrelated body content" and insert specific headings.
Capability inventory: Capability to write to GitHub PRs, upload media via CLI extensions or browser, and delete local files.
Sanitization: Instructions specify redacting secrets (JWTs, tokens) from output and logs to prevent credential leakage.
[DATA_EXFILTRATION]: While the skill's primary purpose is uploading data to GitHub, it handles sensitive authentication material. It implements a worktree-local authentication strategy and includes strict prohibitions against exposing signed URLs, query strings with secrets, or tokens in its outputs.

tk-pr-media