The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents a custom grammar feature that relies on a ".claude-mem.json" configuration file to specify NPM packages (e.g., "tree-sitter-solidity") for dynamic loading. Loading code modules based on paths or package names defined in project-local configuration files represents a dynamic execution risk.- [INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves parsing and displaying the contents of a codebase, which is untrusted input. Malicious instructions embedded in the code could influence the agent's behavior during discovery. (1) Ingestion points: File content read via smart_search, smart_outline, and smart_unfold. (2) Boundary markers: No delimiters or "ignore" instructions are specified to separate code content from agent instructions. (3) Capability inventory: The agent is given structural parsing tools and instructed to use them as its primary discovery method. (4) Sanitization: The skill does not describe any sanitization of the retrieved code content.- [PROMPT_INJECTION]: The skill contains instructions that explicitly override the agent's default behavior, stating "This skill overrides your default exploration behavior" and directing the agent to avoid standard discovery tools like Grep, Glob, and Read.- [EXTERNAL_DOWNLOADS]: The documentation encourages the installation of third-party NPM packages to support custom grammars, introducing external dependencies and potential supply chain vulnerabilities.

smart-explore