The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes vendor/bin/pint for code formatting and php artisan test for running the test suite. It also identifies and runs repository-specific static analysis and focused test commands discovered from the environment.
[PROMPT_INJECTION]: The skill reads work items (tickets and specs) from markdown files located in .scratch/ and issue tracker files. These files are used to determine the scope of implementation, which creates an indirect prompt injection surface where instructions embedded in ticket descriptions or comments could influence agent behavior.
Ingestion points: Reads markdown files under .scratch/ and .agents/issue-tracker.md to scope and claim work.
Boundary markers: Uses markdown headings and frontmatter (e.g., status, blocked-by, Build Contract) to structure data, but does not explicitly instruct the model to ignore potential instructions within the content.
Capability inventory: Subprocess execution (tests, formatters), file system modification (git commit, status updates), and skill chaining (tdd, code-review, verify).
Sanitization: No sanitization of input data from the tracker or scratch files is mentioned.

implement