← ClaudeAtlas

gauntletlisted

Run an independent dual-engine code-review gauntlet (default Claude plus Codex; Grok can substitute for either while it has usage) for a committed branch or PR when the user asks for a gauntlet, adversarial review, or dual-engine review.
Traction-AI-Labs/skills · ★ 1 · Code & Development · score 74
Install: claude install-skill Traction-AI-Labs/skills
# Gauntlet Review one pinned committed diff with two independent reviewers, Claude and Codex by default, and act on what they find. This gauntlet is a verification gate on a candidate-final SHA, never a discovery instrument. Enter only after the branch is frozen. Before round 1, run one independent whole-scope pre-review through this runner (Round 1 gate below) and repair every accepted finding from it in one batch. Every gauntlet round reviews the whole pinned scope, never only the latest delta or repair. ## Review budget This is the default budget (17 Oct 2026, retro ruling O1). A repo's own budget, in its CLAUDE.md or AGENTS.md, overrides it. The orchestrator is the session that owns the PR's merge; a lane running this skill reports to it at the budget and stops. The budget overrides the stop state below, and one run of this skill is one pass over one head: another pass starts only while the budget has a round left. **A round is one reviewed head**, whoever reviews it and whatever it is called: a lane's own review, a bot, a gauntlet round, a targeted or light check. Two reviewers on the same head are one round. The one pre-review before round 1 is part of building the change and is not a round. The runner's round number counts gauntlet heads only, so count the budget from the PR table. | What the change is | Rounds | |---|---| | Docs and add-only tests | 0: the skill does not run | | Plans, SOPs, architecture docs, runbooks, any CLAUDE.md, AGENTS.md or SKILL.md | at m