write-adr
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local commands including
git,ls, andfindto gather repository context and existing ADR numbering. It also calls a local Python scriptplugins/beagle-analysis/skills/adr-writing/scripts/next_adr_number.py. These operations are standard for development workflows and are scoped to the project directory. - [INDIRECT_PROMPT_INJECTION]: The skill processes conversation history to extract architectural decisions, which is a potential surface for indirect prompt injection if the conversation contains malicious instructions.
- Ingestion points: Conversation history analyzed during Step 2 (Extraction).
- Boundary markers: The process enforces a specific JSON schema for extraction results as a structural gate.
- Capability inventory: File writing to
docs/adrs/, execution of repository-level discovery commands, and a local Python script. - Sanitization: A mandatory user confirmation step (Step 3) requires the user to review and manually select decisions before any ADR files are generated or scripts are run for numbering, serving as a robust mitigation.
Audit Metadata