adr
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local environment, including project code, tests, documentation, and history, to synthesize rationale for ADRs. This creates a surface for indirect prompt injection where adversarial content within these files could influence the agent's output.
- Ingestion points: The skill reads decision evidence, code, tests, maintained documentation, and existing ADRs from the local repository (referenced in SKILL.md).
- Boundary markers: The instructions do not define explicit delimiters for separating project content from agent instructions, though they advise the agent to separate facts from supported inferences.
- Capability inventory: The agent has the capability to write and edit files (ADRs) within the project directory (e.g.,
docs/adr/), modify existing records, and update historical links. - Sanitization: No explicit sanitization or filtering of the ingested project content is mentioned in the instructions.
Audit Metadata