The Agent Skills Directory

[SAFE]: The skill demonstrates high privacy standards by strictly prohibiting individual-level output and requiring the pseudonymization of all student data. No evidence of unauthorized network calls, data exfiltration, or hardcoded credentials was found.
[NO_CODE]: The skill consists exclusively of markdown instructions and templates; no scripts, binaries, or other executable files are included, which minimizes the technical attack surface.
[PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection because its primary purpose is to process student-generated text from diagnostics and questionnaires. 1. Ingestion points: Student response data and diagnostics enter the context via file exports provided by the user in the workflow intake phase. 2. Boundary markers: The skill does not define specific technical delimiters (such as XML tags or security markers) for the interpolation of student text. 3. Capability inventory: The agents are limited to reading local files and writing to the course passport; they lack tools for shell command execution or network exfiltration. 4. Sanitization: The skill employs data aggregation and mandatory human-in-the-loop checkpoints (Phase 2), ensuring that individual malicious inputs are filtered through statistical processing before any data is persistently written.

cohort-analyst