The Agent Skills Directory

[COMMAND_EXECUTION]: The skill identifies and executes shell commands such as tests, linting, and building, which are discovered from the project's local configuration files like Makefile, package.json, and CLAUDE.md. While necessary for the skill's purpose, this allows for the execution of arbitrary commands if the repository files are malicious.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection due to its interaction with external and local data sources.\n
Ingestion points: Data enters the agent's context through user-provided arguments, results from web searches, and project configuration files like CLAUDE.md and package.json.\n
Boundary markers: The workflow uses markdown headers to organize phases, but there are no specific markers or instructions to isolate or ignore commands that might be embedded in the ingested data.\n
Capability inventory: The agent can execute shell commands, manage subagents, use a web browser for UI testing, and perform git commits.\n
Sanitization: The skill lacks explicit sanitization or validation mechanisms for the data it ingests from project files or external searches.

implement-feature