The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a validator script (scripts/validate.sh) that uses standard Unix utilities (Bash, Perl, Awk), Ruby, and Git to perform linting and format checks on other skills in a repository. This is intended for local development and maintenance workflows.
[DYNAMIC_CONTEXT_INJECTION]: The skill documentation (references/executable-code.md) and evaluation fixtures describe and utilize the !command syntax. This feature allows shell commands to be executed at skill load time to inject live context (e.g., git describe or git diff) into the agent's prompt. This is a documented feature of the Agent Skills format and the Claude Code platform.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to audit and rewrite existing skill instructions, this skill inherently processes untrusted data (the skills being audited). This creates a surface for indirect prompt injection where a malicious skill could attempt to subvert the auditing agent. The risk is mitigated by the structured audit dimensions and ordered rewrite procedures defined in the references/improving-existing-skills.md file.
[REMOTE_CODE_EXECUTION]: The documentation mentions installing the agent-evals CLI for grading assertions. Both the agent-skills-creator and agent-evals are authored by the same vendor (mblode), and the reference points to the official GitHub repository (github.com/mblode/agent-evals).

agent-skills-creator