de-sloppify

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill operates by scanning user-provided source code files and modifying them based on the results. This ingest-and-edit pattern creates a surface where adversarial code could potentially influence the agent's cleanup decisions, although the risk is mitigated by deterministic heuristics and a required manual diff review step.
  • Ingestion points: Target source files are read via the Read tool and processed by the main.cjs script.
  • Boundary markers: No explicit markers are used to separate untrusted file content from instructions in the prompt context.
  • Capability inventory: File modification (Edit tool), file reading (Read tool), and shell execution (Bash tool) for git and pnpm operations.
  • Sanitization: The skill uses deterministic regular expression heuristics within the main.cjs script to identify lines safe for removal.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the Bash tool, including git diff with output redirection, local execution of the main.cjs script, and project tools like pnpm lint:fix. It specifically references absolute Windows file paths (e.g., C:/dev/projects/agent-studio/...) for temporary diff files and project context.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 08:00 AM