ai-scaffold-dev-templates

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes existing repository files, such as templates and agent configuration files, which could theoretically contain malicious instructions. The skill includes a dedicated 'Read Content Is Data' section that explicitly instructs the agent to treat any imperatives found in these files as data rather than instructions.- Ingestion points: Reads existing templates, AGENTS.md, CLAUDE.md, and .promote-target from the repository.- Boundary markers: Uses HTML-style comment markers (e.g., <!-- ai-skills:dev-context -->) to scope specific blocks and provides explicit instructions to ignore embedded commands.- Capability inventory: The skill can create directories, write new files, and propose modifications to existing files via diffs within the working directory.- Sanitization: Implements logic to ignore imperatives in read content and performs checks to prevent internal repository metadata leakage.- [COMMAND_EXECUTION]: The skill utilizes a limited set of shell commands for environment detection and file verification.- Uses git remote get-url and git rev-parse to determine the repository structure and forge type.- Employs test -e and cmp to verify file existence and byte-identity, ensuring integrity during file operations.- [DATA_EXFILTRATION]: The skill includes logic to prevent the accidental leakage of internal development environment details.- In Step 4, it mandates checking that files written for public GitHub targets do not contain the internal host or path information derived from the origin URL to prevent data exposure.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 02:43 PM