The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages local shell utilities including git, gh, curl, and jq to manage repository state, interact with the GitHub API for project configuration, and perform network requests for research.
[DYNAMIC_EXECUTION]: It performs a dry-run execution of dynamically generated code via scripts/dry-run.sh to validate the structural integrity of a new skill before submitting a Pull Request.
[INDIRECT_PROMPT_INJECTION]: As a meta-programming tool, it uses a natural language string (${var}) to define the logic of a new skill, creating a surface where malicious instructions could influence generated code.
Ingestion points: User-provided description via the ${var} variable in SKILL.md.
Boundary markers: Absent.
Capability inventory: File system writes (skills/), repository modification, and local command execution.
Sanitization: Includes automated checks for YAML validity and scans for development placeholders (e.g., 'TODO', 'your_api_key_here') in the generated skill text.
[DATA_EXPOSURE_AND_EXFILTRATION]: Uses gh api to list the names of existing repository secrets to facilitate the configuration of new skills, though it does not access or expose the values of those secrets.

create-skill