verification-before-completion
Pass
Audited by Gen Agent Trust Hub on Apr 24, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists entirely of instructional text aimed at improving the reliability and honesty of the agent's responses. It does not request sensitive data, initiate network connections, or execute hidden scripts.
- [PROMPT_INJECTION]: The skill uses strong language to enforce behavioral constraints (e.g., 'The Iron Law', 'non-negotiable'). These instructions are intended to prevent hallucinations and ensure accuracy, rather than bypassing security filters or overriding core safety protocols.
- [INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow where the agent must ingest and process external data (terminal output) to verify its work. While this is an intended function, it creates a surface area for indirect instructions if the environment output is maliciously crafted.
- Ingestion points: Command-line output from verification tools (e.g., tests, linters, builds) as specified in SKILL.md.
- Boundary markers: None; the agent is instructed to read the 'FULL command' output without specific delimiters.
- Capability inventory: The agent is expected to execute terminal commands and interpret the resulting logs.
- Sanitization: No explicit sanitization or filtering of command output is defined within the skill instructions.
Audit Metadata