The Agent Skills Directory

[SAFE]: The skill implements a robust defensive architecture against prompt injection. The rules/context-source-safety.md and SKILL.md files explicitly instruct the agent to treat all external data (retrieved text, tool outputs, examples) as 'evidence' rather than 'authority', effectively preventing untrusted content from overriding system instructions.
[SAFE]: The skill includes a comprehensive evaluation suite (assets/evals/prompt-maker-cases.jsonl) that specifically tests for safety, adversarial inputs, and source boundary violations. This ensures that the agent's behavior is verified against potential attack vectors during development.
[SAFE]: The provided validation script (scripts/validate-prompt-maker.mjs) is a benign utility used to enforce project structure and metadata consistency. It relies on standard Node.js/Bun built-in modules (fs, path, process) and performs no network requests or arbitrary command execution.
[SAFE]: Data safety is prioritized through guidelines that forbid the use of global or home directories as authoritative sources and require explicit gating for any credentialed or destructive actions at the prompt boundary.

prompt-maker