The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs analysis on untrusted codebase files including manifests (e.g., package.json, Cargo.toml), configuration files, and source code. 1. Ingestion points: The skill reads various project files across Phases 1, 2, and 3 to identify the tech stack and conventions. 2. Boundary markers: There are no explicit instructions for the agent to ignore prompt-like content or malicious instructions embedded within the analyzed files. 3. Capability inventory: The skill is designed to read codebase files and write/update a CLAUDE.md file in the project root. 4. Sanitization: The instructions do not specify sanitization or validation of the text extracted from the codebase before it is included in the generated reports.
[COMMAND_EXECUTION]: The skill detects common development commands (e.g., build, test, lint) from the codebase and includes them in the onboarding artifacts. While it does not execute these commands directly, it provides them for the user or agent to use, which could carry risk if the analyzed project manifest contains malicious scripts.

codebase-onboarding