The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions require installing the ast-index CLI via a Homebrew tap (defendend/ast-index). This involves downloading a binary tool from a vendor-controlled repository.
[REMOTE_CODE_EXECUTION]: By installing and executing the ast-index binary, the agent runs code provided by the skill author to perform code searches and analysis.
[PERSISTENCE]: The skill includes a utility command, install-claude-plugin, which installs a plugin into the agent's local environment (~/.claude/plugins/). This is a persistence mechanism that allows the skill to modify or extend the agent's behavior across sessions.
[COMMAND_EXECUTION]: The skill facilitates the execution of various CLI commands, including a query command that allows raw SQL to be executed against the project's search index database.
[DYNAMIC_EXECUTION]: The skill uses ast-grep (sg) for structural AST-based code searching (agrep) and provides direct SQL access to the underlying SQLite database.
[INDIRECT_PROMPT_INJECTION]: The skill processes and retrieves content from a user's codebase. This creates an attack surface where maliciously crafted instructions embedded in source code or comments could be interpreted by the agent when searching the project.
Ingestion points: Reads files and symbol definitions across the entire indexed codebase.
Boundary markers: None; content is returned as plain text or structured JSON for the agent to process.
Capability inventory: The agent has CLI access, file system access, and the ability to execute the vendor's binary and SQL queries.
Sanitization: No explicit sanitization or filtering of codebase content is mentioned in the search results.

ast-index