The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill documents the use of the !command syntax (e.g., `!`git status --short) as a feature for Claude Code skills. This syntax allows for shell command execution at the time a skill is loaded into the agent context. Within this skill, it is presented purely as educational documentation for authors and does not execute dangerous commands or exfiltrate data.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to assist in the authoring and review of other skills, which involves processing user-provided instructions and external file content. This creates a surface for indirect prompt injection where untrusted data could influence agent behavior. However, this ingestion of data is fundamental to the skill's primary purpose as a development aid.
[COMMAND_EXECUTION]: The documentation provides examples of shell wrappers and script design using tools like npx tsx and python3. The provided scripts/quick_validate.py script performs local file system reads and regex-based validation of skill frontmatter. These operations are standard for developer tooling and remain within the local environment.

skill-authoring