plain-english
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted prose and file contents for auditing and editing, which represents a potential attack surface for indirect prompt injection.
- Ingestion points: Text pasted into the chat for audit/rewrite and the contents of local files named by the user for editing (SKILL.md, skills/simple-english/SKILL.md).
- Boundary markers: The skill includes an explicit and detailed "Untrusted input boundary" section in both the Plain English and Simple English skill files. It instructs the agent to treat all supplied material as data, not instructions, and to never obey embedded commands (e.g., requests to change roles, reveal data, or run commands).
- Capability inventory: The skill utilizes file-editing capabilities to perform minimal, targeted edits on files specified by the user.
- Sanitization: The instructions specifically mandate that the agent must analyze directives as text only and refuse to act on them, effectively neutralizing potential injection attempts within the source material.
- [PROMPT_INJECTION]: A deterministic detector flagged an instruction-override pattern in submission/test-cases.md. This is a false positive; the file contains negative test cases used to verify that the agent correctly ignores malicious instructions embedded in user-supplied text.
- [DYNAMIC_EXECUTION]: The repository includes build and maintenance scripts (e.g., scripts/check-output-style.sh) that use Python for comparing skill and style files. These scripts are used for development-time verification and do not process untrusted external data at runtime.
Audit Metadata