The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted local data such as git diffs and file contents to determine Pull Request structure and generate commit/PR metadata, which presents a surface for indirect prompt injection.\n
Ingestion points: The skill reads git status, git diff, and recent logs as described in SKILL.md to classify work units.\n
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious instructions embedded within the code or diffs being processed.\n
Capability inventory: The skill possesses significant capabilities including executing shell commands via git, gh, gt, and git-town, as well as writing to the local filesystem (e.g., docs/adr/ and transient body files).\n
Sanitization: The skill logic does not include explicit sanitization or filtering of the ingested diff data before it is used to generate PR descriptions.\n- [EXTERNAL_DOWNLOADS]: The skill references and provides instructions for installing several external developer tools from established sources, including the github/gh-stack extension for the GitHub CLI, git-town, and the Graphite CLI (@withgraphite/graphite-cli).\n- [COMMAND_EXECUTION]: The skill uses various command-line utilities to manage the PR lifecycle. These include git for local operations, gh for GitHub interactions, and stack providers gt and git-town for managing dependent PR chains. These actions are consistent with the skill's intended purpose.

plate