codex-prompt-optimize
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill acts as a meta-utility to improve prompt engineering. It explicitly defines security-positive patterns such as
<action_safety>to prevent scope drift and<verification_loop>to ensure task correctness before finalization. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to transform it into structured instructions. It includes built-in mitigations by requiring clarifying questions for ambiguous inputs and using delimited XML blocks to separate task instructions from metadata.
- Ingestion points: User input is intercepted and analyzed in
SKILL.md. - Boundary markers: The skill wraps processed input into specific XML blocks (
<task>,<default_follow_through_policy>) to provide structural boundaries. - Capability inventory: The skill flow in
SKILL.mdemits an optimized prompt for agent execution. - Sanitization: The skill employs a strict analysis protocol (Intent, Scope, Constraints, Done-State) to normalize and filter vague or conflicting user input into a safer, structured format.
Audit Metadata