implementation-intention-designer

Pass

Audited by Gen Agent Trust Hub on May 14, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill uses instructional markers like 'IMPORTANT' to emphasize specificity in the 'if-then' plans. These are benign and intended for guiding the model's logic, not for bypassing safety filters or overriding system instructions.
  • [DATA_EXFILTRATION]: No network operations (curl, wget, etc.) or access to sensitive local file paths (e.g., .ssh, .aws, credentials) are present in the prompt or metadata.
  • [COMMAND_EXECUTION]: The skill is designed strictly for text generation and does not invoke shell commands, subprocesses, or system-level APIs.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources via variables like student_profiles. While this represents a theoretical ingestion surface for untrusted content, the skill lacks the capability to execute commands or perform network operations, limiting the impact to text-based context manipulation within the LLM session.
  • [OBFUSCATION]: The content is entirely in plain text with no evidence of Base64, hex encoding, zero-width characters, or hidden acrostics.
Audit Metadata
Risk Level
SAFE
Analyzed
May 14, 2026, 03:33 PM
Security Audit — agent-trust-hub — implementation-intention-designer