The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to identify and run project-specific commands (e.g., from package.json or Makefile) to exercise code changes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from the local repository, including git diff output and existing configuration files, which could potentially contain malicious instructions meant to influence the agent's behavior during the verification process.
Ingestion points: git diff, package.json, .agents/skills/*.md, README.
Boundary markers: None specified to distinguish untrusted file content from instructions.
Capability inventory: Shell command execution, web UI automation (Playwright), and file writing.
Sanitization: No explicit sanitization of data ingested from the repository.
[DYNAMIC_EXECUTION]: The skill includes instructions to save successful verification steps as new skill files (.agents/skills/run-<app>/SKILL.md), allowing the agent to persist and reuse complex execution logic generated during a session.

verify-app