instruction-cleanup

Pass

Audited by Gen Agent Trust Hub on Apr 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (touch, rm, grep, printf, cksum, awk) to manage a project-specific flag in the system temporary directory and verify changes.- [PROMPT_INJECTION]: The instructions describe a method to bypass the instruction-guardian reminder mechanism by creating a specific flag file in the system temp directory. This is intended to suppress reminders during the implementation phase.
  • [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted project data:
  • Ingestion points: The skill reads and audits existing project files such as CLAUDE.md, AGENTS.md, and MEMORY.md.
  • Boundary markers: There are no explicit instructions for using delimiters or boundary markers when reading and processing these files.
  • Capability inventory: The skill allows the agent to execute shell commands and write or modify various project files.
  • Sanitization: No sanitization or validation of the content within the ingested instruction files is performed before they are acted upon.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 22, 2026, 07:03 PM