Verify Generator outputs against the Spec with Codex cross-checks.
npx skills add https://github.com/FlineDev/TandemKit --skill evaluator-flinedev
Please help me install this Agent Skill. Skill: evaluator Source: https://github.com/FlineDev/TandemKit/tree/main/skills/evaluator Command: npx skills add https://github.com/FlineDev/TandemKit --skill evaluator-flinedev
Autonomous quality gate for TandemKit outputs: it verifies the Generator’s work against the Spec and leverages Codex as a second opinion to ensure robust evaluation.
Invoke the Evaluator with the session command for your mission, e.g., /tandemkit:evaluator 005-AddDarkMode to begin an autonomous evaluation cycle.
High-intent search queries and answers about installing and using this skill.
To automate code verification against a spec, you can use an autonomous evaluator that cross-checks Generator outputs against Spec.md. It leverages Codex as a second opinion to ensure robust evaluation without requiring a human-in-the-loop.
Codex provides a second opinion for autonomous evaluation by running in parallel to cross-check findings. This cross-verification mechanism prevents rubber-stamping and ensures robust evaluation of Generator results against the Spec.
Generating audit-ready artifacts for multi-round missions is best achieved through a round-based convergence process. This mechanism produces Claude-01.md, Claude-02.md, Codex-01.md, and final Round-NN.md artifacts stored in the Evaluator/Round-NN-Discussion folder.
Yes, you can evaluate generator outputs independently without human intervention. The autonomous evaluator verifies Generator results against Spec.md and leverages Codex for a second opinion to ensure strict evidence for each criterion.
Yes, TandemKit evaluator requires strict evidence for each verification criterion. It reports verdicts with traceable justification and stores all evaluation results and artifacts in the Evaluator/Round-NN-Discussion folder.
To start an autonomous evaluation cycle for a TandemKit mission, invoke the Evaluator with your session command. For example, use /tandemkit:evaluator 005-AddDarkMode to begin coordinating planning, generation, and evaluation cycles.
NAKKA-KDefine rubrics, run baseline evaluations, and compare before-and-after outcomes.
sliorEvaluate convergence across agent refinements and generate final synthesis documents.
EvoClawCoordinate iterative multi-agent discussions to converge on a resolved artifact.
duongtdnReview GitHub issues and verify implementations against acceptance criteria.
