The Agent Skills Directory

[SAFE]: The skill implements significant safety protocols for mental health assessment, including automatic crisis assessment triggers (based on PHQ-9 Question 9) and mandatory crisis resource provision. It explicitly prohibits providing medical diagnoses or drug dosages.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted user data stored in local JSON files.
Ingestion points: The agent reads user-provided text from data/psych-assessments/*.json (e.g., baseline.reason_for_assessment) to generate reports and conduct support dialogues.
Boundary markers: The instructions lack specific delimiters or "ignore instructions" warnings for the agent when it re-processes this previously saved data.
Capability inventory: The skill uses Read, Write, Glob, and Grep tools to manage these files.
Sanitization: No explicit sanitization or filtering logic is defined for data read from the local assessment database. However, the risk is localized as the skill lacks network access to exfiltrate session data.

psych-assess