The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external source code (Ruby/Rails files) provided by the user to identify structural patterns. This creates a surface for indirect prompt injection if the source code contains malicious instructions meant to influence the agent. However, the skill includes a mandatory 'HARD-GATE' and 'SECRET SAFETY' rule set that explicitly instructs the agent to verify findings and redact any sensitive credentials (API keys, tokens, etc.) found in the input files, which mitigates standard injection and data exposure risks.
[CREDENTIALS_UNSAFE]: The skill explicitly includes logic in SKILL.md to detect and redact sensitive data. It forbids the reproduction of secrets, tokens, and passwords in its output, using a redacted fingerprint instead. This follows security best practices for handling sensitive context.

review-architecture