cognitive-walkthrough

Pass

Audited by Gen Agent Trust Hub on May 19, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the user ($ARGUMENTS) and the project's codebase context without using explicit boundary markers or delimiters. A malicious user or an attacker who can modify the codebase could embed instructions that redirect the agent's behavior during the walkthrough process.
  • Ingestion points: $ARGUMENTS variable and codebase context referenced in SKILL.md.
  • Boundary markers: None present to isolate instructions from data.
  • Capability inventory: The agent has the ability to create directories and write files (docs/hci/cognitive-walkthrough.md).
  • Sanitization: No input validation or instruction filtering is implemented.
  • [COMMAND_EXECUTION]: The skill instructs the agent to interact with the file system by creating directories and writing a markdown report to docs/hci/cognitive-walkthrough.md. While these are intended features for documenting the UX analysis, they represent a capability that could be misused if the prompt logic were subverted via injection.
Audit Metadata
Risk Level
SAFE
Analyzed
May 19, 2026, 04:52 PM
Security Audit — agent-trust-hub — cognitive-walkthrough