to-spec
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the conversation history and the codebase to generate documentation, creating a surface for indirect prompt injection.
- Ingestion points: The agent is instructed to use "el contexto de la conversación actual" (current conversation context) and "conocimiento del codebase" (codebase knowledge) as primary inputs (SKILL.md).
- Boundary markers: Absent. There are no instructions provided to the agent to sanitize, escape, or use delimiters when incorporating external content into the final specification document.
- Capability inventory: The skill has the capability to write files to the repository (
docs/specs/or.hermes/plans/) and publish content to GitHub as issues (SKILL.md). - Sanitization: Absent. The process "sintetiza lo que ya sabes" (synthesize what you already know) does not include validation steps for the content being synthesized.
- Downstream impact: The mention of a
ready-for-agentlabel for GitHub issues suggests that the generated output is intended to be consumed by other automated agents, increasing the risk that a successfully injected instruction could be executed by a downstream system.
Audit Metadata