The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various external sources, including mood diaries and trigger logs, which creates a surface for indirect prompt injection.
Ingestion points: The agent reads user-generated content from data-example/mental-health-logs/*.json and data-example/mental-health-tracker.json to identify patterns and triggers.
Boundary markers: The instructions lack technical delimiters or explicit safety instructions to prevent the agent from obeying instructions that might be embedded within the user's diary entries or log data.
Capability inventory: The skill uses Read, Grep, Glob, Write, and Edit tools. It specifically generates interactive HTML reports using the Write tool.
Sanitization: There is no evidence of sanitization or escaping of the user-provided strings before they are interpolated into the final analysis or the generated HTML reports, which could allow malicious content to influence the agent's logic.
[DATA_EXFILTRATION]: The skill is designed to handle highly sensitive Personal Health Information (PHI), including depression and anxiety scores (PHQ-9/GAD-7). While no network-based exfiltration was detected, the skill moves this sensitive data into generated HTML reports using the Write tool. This transformation of raw data into a portable format increases the risk of data exposure if the reports are shared or stored insecurely.
[COMMAND_EXECUTION]: The skill generates interactive HTML reports (Example 6) that incorporate user-supplied text from mental health logs. If an attacker embeds malicious HTML or JavaScript (XSS) within a diary entry, the agent may unknowingly write it into the generated report. If the user then opens this report in a browser, the scripts could execute in the user's local context.

mental-health-analyzer