ai-scaffold-agents-md
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes content from
AGENTS.mdandCLAUDE.mdfiles which could contain untrusted instructions. It correctly mitigates this risk through a 'Read Content Is Data' policy, instructing the agent to move content verbatim and ignore any imperatives found within the files. Findings: Ingestion points: Reads local context files; Boundary markers: Explicit instructions to treat imperatives as data; Capability inventory: Local file operations only; Sanitization: Post-move verification withcmp. - [COMMAND_EXECUTION]: Uses standard local commands (
git,test,cmp) to manage files. The scope is limited to the repository root, with explicit validation logic (Edit Scope) to prevent directory traversal or access to sensitive system paths like/etcor home directories. - [DATA_EXPOSURE]: File access is restricted to the working directory. The skill explicitly excludes sensitive locations like
.claude/or git-ignored files. No network capabilities are requested or used, preventing data exfiltration risks.
Audit Metadata