verification-before-completion

Pass

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses extremely forceful instructions and behavioral mandates (e.g., 'The Iron Law', 'Non-negotiable', 'Lying') to override agent tendencies to claim success without evidence. While intended for accuracy, this constitutes a pattern of instruction overriding.\n- [PROMPT_INJECTION]: Indirect Prompt Injection Surface (Category 8): The skill directs the agent to execute shell commands and read their 'Full output' to verify its work. This creates a vulnerability where malicious code in a project could produce command output containing instructions that manipulate the agent's behavior.\n
  • Ingestion points: Shell command outputs from tests, builds, and linters as referenced in the SKILL.md documentation.\n
  • Boundary markers: None present in the instructions to delimit command output from agent instructions.\n
  • Capability inventory: The agent is expected to execute subprocesses (tests, builds, linters) and read their output into its context.\n
  • Sanitization: No filtering or sanitization of command output is defined in the instructions.\n- [COMMAND_EXECUTION]: The skill's primary function requires the agent to run various shell commands to verify project state. This grants the agent the ability to execute arbitrary verification scripts or tools found within the environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 29, 2026, 02:15 AM