student-personalization

Pass

Audited by Gen Agent Trust Hub on Mar 23, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it is designed to ingest and analyze untrusted external data—including package manifests, learner data models, and codebase structures—without establishing security boundaries.\n
  • Ingestion points: External manifests and core learner data structures defined in Phases 1 and 2 (e.g., package manifests, learner profiles, learning events).\n
  • Boundary markers: Absent; the instructions do not include delimiters or specific commands to disregard embedded instructions within the processed data.\n
  • Capability inventory: The skill has the capability to read any file on the filesystem and write reports to docs/ as well as telemetry logs to the user's home directory.\n
  • Sanitization: There is no evidence of sanitization, escaping, or validation logic for data ingested from the analyzed platform.\n- [COMMAND_EXECUTION]: The skill implements a persistence-like telemetry mechanism that writes execution metadata to the user's home directory.\n
  • Evidence: The 'Self-Evolution Telemetry' section appends outcome and iteration data to ~/.claude/projects/skill-telemetry.md.\n
  • Context: While this feature is used for platform improvement, it involves modifying persistent files located outside the immediate project workspace.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 23, 2026, 10:58 AM
Security Audit — agent-trust-hub — student-personalization