The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from web searches and sibling repositories and incorporates it into project files. * Ingestion points: Sibling repositories (../), web search results, and external documentation via MCP. * Boundary markers: None explicitly defined to isolate or distinguish external content in generated documents. * Capability inventory: File-write access to CLAUDE.md, docs/research/, and docs/plans/. * Sanitization: No specific sanitization or validation of external content before writing to the filesystem is mentioned.
[DATA_EXFILTRATION]: The skill instructions specify checking sibling repositories in the parent directory (../). This extends the agent's data access scope beyond the immediate project root to find reference implementations.
[COMMAND_EXECUTION]: The skill performs automated filesystem exploration, including reading build configuration files (package.json, go.mod, Cargo.toml, pyproject.toml) and project structure to detect programming languages and project orientation.

research-plan