The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation in anti-patterns.md (Rule #27) contains keywords associated with prompt injection, such as "ignore previous instructions" and "you are now X". These are used exclusively in a defensive context, instructing the agent to treat such strings in external data as untrusted and to report them to the user rather than obeying them.
[INDIRECT_PROMPT_INJECTION]: The skill defines a "Skill Entry Protocol" in entry-protocol.md and behavioral constraints in anti-patterns.md that establish security boundaries for data ingestion. It explicitly mandates the sanitization of external content and the use of completion criteria to prevent accidental execution of embedded instructions.
[SAFE]: All files provided are Markdown-based documentation and behavioral rules. There is no evidence of obfuscation, remote code execution, persistence mechanisms, or credential harvesting. The skill's configuration properly uses disable-model-invocation: true to limit its scope as a shared resource bundle.

rules