ai-coding-agents-safety-envelope
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is an educational and technical resource for implementing safety boundaries in AI agents. It does not contain any executable code that performs malicious actions.
- [SAFE]: All references to sensitive file paths (e.g.,
~/.ssh/,~/.aws/credentials) and dangerous commands (e.g.,rm -rf,force-push) are provided within security checklists, audit guides, and test matrices designed to help developers identify and mitigate risks. - [SAFE]: External URLs and resource links target official documentation from trusted organizations such as Anthropic, OpenAI, the Linux Foundation (Agentic AI Foundation), and OWASP. These links are documented neutrally and follow standard security practices.
- [SAFE]: The skill provides explicit guidance on avoiding the persistence of secrets and credentials in policy files, recommending standard practices like the use of environment variables and automated token rotation.
- [SAFE]: The analysis of execution surfaces (Category 8) shows that the skill clearly identifies ingestion points (tool arguments), boundary markers (sandbox roots and permission modes), and capabilities (network and filesystem access) for the purpose of defining enforcement policies rather than exploiting them.
Audit Metadata