Валидация технических решений по актуальным практикам

engineering

validate-agent is a Claude Code skill that checks every technology decision in an implementation plan against 2024–2025 best practices before a single line of code is written. Its process has three stages: querying the Artifact Index via a RAG-Judge script (`braintrust_analyze.py`) to surface patterns from past succeeded and failed handoffs; running WebSearch for each tech choice using queries like "best practices 2024 2025" and "deprecated OR recommended"; and assigning each choice a status of VALID, OUTDATED, DEPRECATED, RISKY, or UNKNOWN. Results are written to a structured `validation-<plan-name>.md` handoff file, and the agent returns a top-level verdict of VALIDATED or NEEDS REVIEW to the orchestrator, listing specific issues and suggested alternatives. Teams use it to catch outdated dependencies, security risks, and architectural anti-patterns automatically during the planning phase.