The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the git and gh (GitHub CLI) tools to manage repository state, fetch remote references, and inspect task currency.
[DYNAMIC_EXECUTION]: The file references/readiness-card.md defines and executes a shell function create_verified_snapshot. This function dynamically creates a temporary Git index and interacts directly with the Git object database using git hash-object, git update-index, and git write-tree to produce an isolated content snapshot of the worktree changes.
[INDIRECT_PROMPT_INJECTION]: The verification process involves reading and analyzing untrusted data from the local file system (worktree content, Git diffs) and external platforms (PR metadata), creating a surface for indirect injection.
Ingestion points: Untrusted data enters the agent context through GitHub CLI output in SKILL.md and through Git diffs and file content reads in references/readiness-card.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore embedded instructions within the ingested data.
Capability inventory: The skill possesses capabilities for broad shell execution (Git and CLI operations) and file system manipulation across both files.
Sanitization: The scripts employ defensive shell programming practices, such as proper variable quoting and the use of the double-dash (--) prefix in commands, to prevent argument injection from malicious filenames.

done