The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements PostToolUse hooks in the frontmatter that execute validation scripts (plugin-validate.py and validate-skill-paths.py) using the uv run python3 command. These are standard developer tools used for ensuring plugin integrity.
[DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic bang-backtick syntax (!cat ...) to inject the contents of local reference files (creator-anti-patterns.md and enforcement-checklist.md) into the prompt at load time. This behavior is used to provide the AI with project-specific guidelines and is considered safe for local files.
[INDIRECT_PROMPT_INJECTION]: The skill presents an ingestion surface by reading external reference files via shell commands and interpolating them into the agent's context.
Ingestion points: SKILL.md (via !cat commands targeting files in ../../references/).
Boundary markers: Absent; the injected content is not wrapped in explicit delimiters to differentiate it from the skill's instructions.
Capability inventory: SKILL.md (includes PostToolUse hooks with shell execution capabilities and Skill() tool invocation).
Sanitization: Absent; the skill does not verify or sanitize the content of the reference files before injection.

plugin-creator