The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from GitHub Pull Requests, including titles, descriptions, review comments, and code diffs, which are controlled by external contributors.
Ingestion points: Untrusted data is retrieved in scripts/stack-map.sh using gh pr list and gh api, and further inspected using git show and git diff as described in SKILL.md.
Boundary markers: The instructions do not define clear delimiters or specific instructions for the agent to ignore potentially malicious content embedded within the PR data.
Capability inventory: The skill's capabilities are limited to read-only operations including git fetch, git show, git diff, gh pr list, and gh api calls.
Sanitization: There is no evidence of content sanitization or validation performed on the PR data before it is integrated into the agent's context.
[DYNAMIC_EXECUTION]: The scripts/stack-map.sh script executes an embedded Python script to process JSON data.
Evidence: The script uses a heredoc pattern (`python3
... <<'PY'`) to run Python code directly from the bash script.
Pattern: This execution involves a static, predefined template for data manipulation, which is a common developer practice and poses a low security risk.
[COMMAND_EXECUTION]: The skill involves the execution of various system commands to interact with the repository and GitHub CLI.
Evidence: It invokes git fetch, git show, git diff, gh pr list, gh api, and gh repo view.
Scope: These commands are used within their intended functional scope to retrieve repository metadata and inspect file contents.
[EXTERNAL_DOWNLOADS]: The skill fetches data from GitHub's official services.
Evidence: It uses gh pr list, gh api, and git fetch to retrieve pull request metadata and update remote-tracking branches from the origin repository.

stack-sweep