enhance-skill-prompts
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and modify external SKILL.md files, which introduces a surface for indirect prompt injection where malicious instructions in the input file could attempt to subvert the enhancement process. 1. Ingestion points: The agent reads target SKILL.md files during Phase 0 and Phase 1 to analyze their structure, register, and content. 2. Boundary markers: The skill explicitly defines a Phase 2 invariant section that mandates the agent maintain the original name, description triggers, scope, behavior, and stance, serving as a critical guardrail against malicious overrides. 3. Capability inventory: The skill is restricted to reading local files and writing modifications to markdown files; it does not utilize network operations, remote script execution, or sensitive system APIs. 4. Sanitization: The skill relies on a mandatory self-critique rubric and invariant validation to ensure that modifications do not diverge from the original skill's intended purpose or introduce unintended logic.
Audit Metadata