The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates as a factory that ingests user descriptions via AskUserQuestion to generate new instruction sets in the form of SKILL.md files. While it includes mandatory human review checkpoints before writing files, it lacks automated sanitization for the generated instruction logic, creating a surface where malicious instructions could be embedded in the resulting skill files.
Ingestion points: User input collected via AskUserQuestion in Step 1 and Step 2.
Boundary markers: Interactive review steps (Step 4 and Step 5b) require explicit user confirmation before finalizing the generated content.
Capability inventory: The skill possesses Write access to the file system and Bash access to the git and gh CLI tools for repository management.
Sanitization: Validates skill names for kebab-case formatting and required prefixes, but relies on human review for the safety and correctness of the generated markdown instructions.
[DYNAMIC_EXECUTION]: The skill generates and writes new executable agent instructions (SKILL.md) to the file system at runtime. This process involves assembling markdown templates with user-provided logic and parameters.
[COMMAND_EXECUTION]: The skill utilizes git and GitHub CLI (gh) tools to automate development workflows, including creating branches, committing generated files, pushing changes to remote repositories, and opening pull requests. These actions are triggered based on user consent during the interactive session.

create-skill