The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process CLAUDE.md files from a repository. Since these files are intended to provide instructions and context to AI agents, they represent an attack surface where adversarial content or malformed instructions could potentially influence the agent's behavior during the audit or improvement process.
Ingestion points: The skill reads CLAUDE.md, .claude.md, .claude.local.md, and ~/.claude/CLAUDE.md as part of its discovery workflow.
Boundary markers: The instructions do not currently include explicit boundary markers or "ignore embedded instructions" warnings when the agent reads external file content.
Capability inventory: The skill utilizes the Edit tool for modifying files and the Bash tool for finding files.
Sanitization: There are no documented sanitization or validation steps for content read from these files before they are analyzed.
System Configuration Access: The skill is designed to interact with and potentially modify configuration files located in the user's home directory (~/.claude/CLAUDE.md), which are used for global agent preferences across projects.
Command Execution: The skill employs standard shell utilities to identify relevant files within the codebase.
Evidence: Execution of a find command piped to head in Phase 1 of the workflow to locate markdown files.

claude-md-improver