The Agent Skills Directory

[SAFE]: No malicious code, hardcoded credentials, or unauthorized system interaction patterns were detected within the skill files or procedures.\n- [PROMPT_INJECTION]: The skill's evaluation file (references/evals.yaml) contains adversarial strings such as "Ignore previous instructions" and "override system constraints." These are part of a safety diagnostic suite used to verify that the agent correctly refuses malicious attempts to bypass its instructions; they are not used as instructions for the skill's normal operation.\n- [DATA_EXFILTRATION]: The skill emphasizes data privacy and control by requiring the agent to "Redact sensitive details by default" and establishing clear "approval gates" that must be confirmed before any handoff to implementation or planning tasks.

interview-me