The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves fetching and acting upon untrusted data from GitHub PR review comments, creating an attack surface where an external reviewer could influence the agent's behavior.
Ingestion points: The skill retrieves PR review threads and comment bodies via gh api graphql in Step 3 and Step 7.3.
Boundary markers: The instructions do not define clear delimiters or provide safety warnings to the agent to disregard instructions potentially contained within the retrieved comment text.
Capability inventory: The skill possesses significant capabilities, including local file system interaction (TodoWrite, Read file), GitHub API mutations (posting comments, resolving threads), and repository modifications (git commit, git push).
Sanitization: No sanitization or validation logic is specified for the comment content before it is processed by the agent to categorize feedback and determine subsequent actions.
[COMMAND_EXECUTION]: The skill heavily relies on shell command execution to perform its logic.
Evidence: Extensive use of the GitHub CLI (gh), git, and shell utilities like sed and jq for parsing arguments and interacting with the remote repository.
[PRIVILEGE_ESCALATION]: The skill's documentation contains instructions involving elevated privileges.
Evidence: In Step 1, the installation guide intended for the user suggests the use of sudo apt install gh to install the required GitHub CLI dependency.

pr-review