learn-from-agent-conversations
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions are focused on providing educational content and explanations. It does not include any executable code, remote script downloads, or network operations.
- [DATA_EXFILTRATION]: The skill includes positive security practices, specifically instructing the agent to remove personal identifiers, machine paths, and credentials from generated learning materials in references/learning-notes.md.
- [COMMAND_EXECUTION]: The skill explicitly distinguishes between explaining a command and executing it, stating that an explanation request does not authorize environment modification or execution in SKILL.md.
- [INDIRECT_PROMPT_INJECTION]: While the skill processes user-provided content such as code snippets or PR bodies, it lacks any tools or capabilities like shell access or network requests that could be exploited via injection. The instructions emphasize sticking to the available content and distinguishing between evidence and hypothesis.
Audit Metadata