The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a wide range of shell commands to automate development tasks. This includes Git operations (commit, push, merge, checkout), GitButler operations for workspace management (but snapshot, teardown, setup), and PR interactions using a local gh-tool utility. These commands are chained to minimize user interruption while maintaining workflow flow.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to fetch and automatically process instructions contained in external PR review comments.
Ingestion points: Fetches untrusted data from GitHub via bun run gh-tool pr threads and bun run gh-tool pr issue-comments-latest (SKILL.md).
Boundary markers: The instructions do not define delimiters or specific "ignore external instructions" guardrails for the fetched comment text.
Capability inventory: The skill has file-write capabilities, can execute the project's validation suite (bun run check), and can push code changes back to the remote repository (SKILL.md).
Sanitization: No content sanitization or validation of the fetched comment strings is described prior to the agent evaluating them for code changes.

git-workflow