sf-eval-starter

What problem does it solve?

This Skill automates the RED→GREEN→REFACTOR evaluation cycle by dispatching SpecFormula MCP jobs, preparing a verifiable test command, and polling for verdicts until completion.

Core Features & Use Cases

• SF MCP goal orchestration: Fetches the next TDD goal and loops until the evaluator returns the all-done sentinel.
• Sub-skill delegation by phase: Delegates phase-specific execution to aibdd-red-execute / aibdd-green-execute / aibdd-refactor-execute.
• Deterministic evaluator-ready evidence: Builds an explicit, non-globbbed test command that includes the current goal feature plus all DONE feature files.
• Runner launch preflight: Dry-runs the test command locally in the worker context and fails over cleanly when the environment can’t launch the runner.
• Automated polling cadence: Polls mcp__sf__verify_goal_status using bash sleep 20 until completed=true and records the final verdict.

Quick Start

Ask your agent to run the sf-eval-starter skill with the /sf-eval-starter trigger; it will enter SF full-auto mode and drive the RED→GREEN→REFACTOR loop via SpecFormula MCP until all goals are verified.