correct

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the modification of core instruction files (such as AGENTS.md, documentation in .context/, and Knowledge Base pages) by analyzing historical session logs and user corrections. This creates a persistence mechanism for instructions found in data, which could be used to bypass existing safety rules or inject malicious behavior into the agent's long-term configuration.
  • Ingestion points: Processes session data, user corrections, and project documentation via the @session-miner tool and direct file reads.
  • Capability inventory: Performs file-write and deletion operations on documentation and instruction files. It also proposes modifications to the codebase.
  • Boundary markers: The instructions do not provide clear boundaries or validation steps to prevent the agent from inadvertently deleting security-critical rules while "thinning" documentation or reacting to a "correction."
  • Sanitization: Lacks a specified process for sanitizing input from historical sessions before incorporating it into the project's permanent instruction set.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 05:09 AM