The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in references/detailed-guide.md direct the agent to perform extensive shell operations including directory creation (mkdir), file content modification using sed, and symbolic link creation (ln). The use of sed for string interpolation is potentially risky if user-provided variables containing special characters are not properly escaped by the agent during execution.\n- [INDIRECT_PROMPT_INJECTION]: The skill acts as a scaffolding engine that interpolates raw user input into newly created skill files, creating an injection surface.\n
Ingestion points: User brainstorming responses captured in Phase 1 (e.g., skill purpose and description).\n
Boundary markers: No clear delimiters or safety instructions are used to separate user-provided text from the generated skill instructions.\n
Capability inventory: The skill has the capability to write to the filesystem (sed) and install skills system-wide (ln).\n
Sanitization: The skill does not perform any automated sanitization or validation of user input before placing it into the SKILL.md or README.md files of the new skill.\n- [PERSISTENCE]: The skill automates the "Global Installation" of newly created skills by creating persistent symbolic links in global configuration paths such as ~/.copilot/skills/ and ~/.claude/skills/.\n- [DYNAMIC_EXECUTION]: The scripts/init_skill.py utility generates a boilerplate Python script and explicitly modifies its file permissions using chmod 0o755 to make it executable.\n- [DATA_EXFILTRATION]: The discovery phase in references/detailed-guide.md instructs the agent to read local git configuration (user.name and user.email) to automatically populate metadata fields in the generated skill files.

skill-creator