The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill implements a structured, security-conscious workflow for developing other agent skills with evidence-backed requirements.
[COMMAND_EXECUTION]: The skill executes local commands using bun (for the provided scaffolding script) and mise (for documentation and linting tasks). These tools are used for repository maintenance and are invoked locally without downloading external scripts.
[DYNAMIC_CONTEXT_INJECTION]: The included scaffold-skill.ts script explicitly scans for and blocks the creation of skills that use dynamic context injection (shell commands wrapped in backticks with a leading exclamation mark) in their templates, preventing the introduction of pre-execution vulnerabilities in generated skills.
[DATA_EXFILTRATION]: Instructions in SKILL.md and references/runtime-trust.md mandate strict secret handling, explicitly forbidding the extraction, logging, or reproduction of credentials found in data or environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill design treats all repository and web content as untrusted data and includes clear instructions to the agent that embedded directives must not be allowed to bypass existing safety rules or task scope.

tailrocks-skill-create