The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands including find, git log, and head to collect metadata and file snippets from the local environment. It also executes a vendor-provided script, grep-api.js, using Node.js to interface with the research service.
[DATA_EXFILTRATION]: Project-specific data—including directory structures, recent git history, and the contents of files like CLAUDE.md and dependency manifests—is consolidated into a temporary file and sent to an external research service. This data transmission is the central functionality of the skill.
[INDIRECT_PROMPT_INJECTION]: By ingesting and processing content from files within the local codebase to inform its research, the skill creates an attack surface where malicious instructions embedded in the codebase could manipulate the research output or the agent's subsequent implementation plans.
Ingestion points: The skill reads CLAUDE.md, package.json, pyproject.toml, requirements.txt, Cargo.toml, go.mod, and arbitrary source files mentioned in the user request.
Boundary markers: No specific delimiters or "ignore instructions" warnings are applied to the codebase content before it is processed.
Capability inventory: The skill can read local files, execute shell commands, and perform network requests via the grep-api.js script.
Sanitization: The skill does not perform sanitization, filtering, or validation of the codebase content gathered for the research context.

grep-plan