The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands to inspect the repository and validate the build.
In SKILL.md, the agent is instructed to run git status, git log, and git diff to evaluate changes.
The skill requires running the project's test suite and executing the application or CLI (e.g., 'run the app, hit the endpoint') for manual validation.
Git write operations are performed, including git commit and git push to the main branch.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and acts upon untrusted data from the workspace.
Ingestion points: The agent reads git diff content, git log messages, and the output of a 'review subagent'.
Capability inventory: The agent has the authority to write files (to 'fix' findings), execute arbitrary shell commands (during 'validation'), and push code to remote repositories.
Boundary markers: The instructions do not define delimiters or specific 'ignore instructions' warnings for the content being reviewed in the diffs or subagent feedback.
Sanitization: No sanitization or escaping of the diff content or subagent findings is performed before the orchestrator decides to 'Fix findings'.

ship