Confidence Check

The Confidence Check skill acts as a rigorous gatekeeper for software engineering workflows, specifically designed for agents and developers using Claude Code. By mandating a confidence score of 90% or higher before initiating any implementation task, it effectively prevents premature coding, reduces technical debt, and minimizes 'wrong-direction' execution. This tool orchestrates a systematic verification process that forces the agent to validate the technical feasibility and strategic alignment of a task against the existing codebase and project standards. It is intended for software engineers, systems architects, and AI agent orchestrators who require strict adherence to maintainability and architectural integrity.

• Performs automated duplicate functionality detection using advanced Grep and Glob search patterns across the repository.

• Validates architecture compliance by cross-referencing CLAUDE.md and PLANNING.md guidelines to ensure stack alignment with tools like Supabase, UV, and pytest.

• Enforces strict review of official documentation via Context7 MCP and WebFetch to verify API compatibility and avoid reliance on outdated assumptions.

• Identifies proven OSS implementations through Tavily MCP and web search integration to provide baseline references.

• Conducts root cause identification by analyzing logs, stack traces, and error messages to ensure the underlying problem is solved rather than just patching symptoms.

• Implements a weighted scoring system (0.0 to 1.0) where each check contributes to a final readiness total, providing a clear pass/fail signal for development readiness.

• Always run this check before opening a new file or modifying core modules to ensure your changes are verified and necessary.

• The tool expects structured input and provides an output summary formatted for easy review, indicating whether to proceed, refine alternatives, or request more context.

• Designed as a meta-programming configuration; while the TypeScript reference is in confidence.ts, the skill operates as an instruction-injected workflow.

• Expected output includes a granular breakdown of the five criteria and a final percentage score to justify the development decision.

• Constraints: Requires consistent project documentation (CLAUDE.md, PLANNING.md) to function optimally; relies on external network tools for real-time documentation and OSS verification.