architecture-decision-records
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads existing ADR files from the
docs/adr/directory to answer user questions about architectural choices. This ingestion of potentially collaborator-controlled content creates a surface for indirect prompt injection if an ADR contains malicious instructions. - Ingestion points: Reads content from
docs/adr/README.mdand individual ADR markdown files. - Capability inventory: The skill has the ability to create directories and write new markdown files to the local file system.
- Boundary markers: The instructions do not explicitly tell the agent to ignore any natural language instructions that might be embedded within the files it reads.
- [COMMAND_EXECUTION]: The skill performs file system management tasks, specifically creating directories and writing new markdown documents.
- Evidence: The workflow includes steps to initialize the
docs/adr/folder and write new decision records (e.g.,0001-use-nextjs.md). - Protection: The skill explicitly instructs the agent to ask for user confirmation before creating the directory and before writing any files, ensuring the user remains in control of file system changes.
Audit Metadata