The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the codebase being analyzed.
Ingestion points: File reading via Read and Grep tools, as well as directory traversal using ls, find, and librarian agent.
Boundary markers: None identified; the agent processes file content directly without specific delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has access to Bash (shell execution) and Edit (file writing).
Sanitization: No evidence of sanitization or filtering of codebase content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform codebase inventory and exploration.
Usage: Automated use of ls, find, and grep via shell to build a module map.
Potential risk: While used for discovery, the presence of the Bash tool with user-influenced file paths or content presents a broad capability surface.

improve-codebase-architecture