The Agent Skills Directory

[COMMAND_EXECUTION]: The skill configures an automated PostToolUse hook in .claude/settings.json that executes a bash script (.claude/scripts/post-edit-check.sh) whenever the agent performs file edits. This creates a persistent execution loop triggered by common agent actions.
[COMMAND_EXECUTION]: The skill utilizes shell commands (cp and chmod +x) to deploy and enable an executable script within the user's project directory.
[PROMPT_INJECTION]: The skill instructions include an indirect prompt injection surface by consuming untrusted data from the project root (e.g., README.md, package.json) to populate the CLAUDE.md system prompt file.
Ingestion points: Files like README.md, package.json, and the overall project directory structure are read to extract descriptions and metadata.
Boundary markers: The skill lacks explicit instructions or delimiters in the generated CLAUDE.md to prevent the agent from following instructions that might be embedded in the ingested project data.
Capability inventory: The skill has the capability to write to the file system, create directories, modify agent settings, and execute shell scripts via hooks.
Sanitization: There is no evidence of sanitization or validation of the content read from project files before it is interpolated into the configuration templates.

Project Scaffolding