thinking-practice

Pass

Audited by Gen Agent Trust Hub on Aug 3, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted data which could contain malicious instructions designed to influence the agent's reasoning or tool use.
  • Ingestion points: Processes untrusted inputs through 'Pasted content' (line 123) and 'Evidence return' (line 186) as defined in SKILL.md.
  • Boundary markers: Implements 'claim hygiene' rules (line 137) to distinguish between user-supplied source claims and verified evidence.
  • Capability inventory: The skill can invoke an external 'research' sub-skill (line 167) and read project metadata from 'docs/agents/project.md' (line 108).
  • Sanitization: Employs field-level provenance (line 231) within its hand-off packets to identify the origin of all data points, helping to prevent the promotion of injected content into decisions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 3, 2026, 09:02 AM
Security Audit — agent-trust-hub — thinking-practice