The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute various shell scripts located in the target repository, such as ./check_codebase.sh, ./build_github_pages.sh, ./run_web_server.sh, and ./run_playwright_tests.sh. This allows the agent to run arbitrary code provided by the project being analyzed.
[COMMAND_EXECUTION]: Instructions include executing Python scripts after sourcing a local environment script (source source_me.sh && python3 ...). This pattern grants the agent the capability to execute potentially malicious logic contained within the repository's Python helpers.
[PROMPT_INJECTION]: The workflow mandates reading and following instructions from AGENTS.md, CLAUDE.md, and docs/TYPESCRIPT_STYLE.md within the target repository. This creates an attack surface for indirect prompt injection, where malicious instructions embedded in these documentation files could manipulate the agent's behavior.
Ingestion points: AGENTS.md, CLAUDE.md, and docs/TYPESCRIPT_STYLE.md (referenced in SKILL.md).
Boundary markers: None identified; the agent is instructed to adopt these local instructions as a "source of truth" for exceptions.
Capability inventory: Shell script execution, Python execution, and file system interaction (documented in SKILL.md).
Sanitization: No sanitization or validation of the ingested file contents is specified.

typescript-engineer