package-optimizer
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements strong defensive programming patterns, including strict evidence validation logic and explicit guardrails such as 'Never edit without explicit approval' which limit the agent's autonomous behavior.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of 'Current evidence' and 'Capability profiles' to generate optimization proposals, creating a surface for indirect injection.
- Ingestion points: Defined in SKILL.md under 'Required inputs' (evidence, profile).
- Boundary markers: Absent; no specific delimiters or ignore-instructions are used for external data.
- Capability inventory: The skill is restricted to text generation; it is explicitly forbidden from editing files, changing profiles, or running benchmarks.
- Sanitization: Absent; the skill does not instruct the agent to sanitize or escape input data.
Audit Metadata