The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to update the CLAUDE.md file, which serves as the primary context for the AI agent in subsequent sessions. This creates a surface for indirect prompt injection where malicious instructions placed in the codebase (e.g., in a Makefile, package.json, or as code comments/conventions) could be automatically incorporated into the agent's long-term memory, potentially influencing future behavior.
Ingestion points: Project configuration files (Makefile, package.json), source code files (*.go, *.py, *.tsx), and directory structures.
Boundary markers: None specified for the content appended to CLAUDE.md.
Capability inventory: File system read/write access to CLAUDE.md and project files; shell execution for auditing commands.
Sanitization: The skill instructions do not specify sanitization or validation of the content being moved from the codebase into the CLAUDE.md file.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for staleness detection, specifically using grep and ls to verify the existence of commands, environment variables, and directories. These are standard project discovery operations but involve executing logic based on strings retrieved from documentation and the file system.

claude-md