← ClaudeAtlas

convergelisted

Use when you want to drive an artifact to clean, where clean means zero blockers and zero majors after a full check run. Fix and re-run until clean or a round cap, then report honestly, confirmed minors included. A fix-until-clean loop over lint, tests, review, and scrutiny.
alawein/outpost · ★ 1 · AI & Automation · score 74
Install: claude install-skill alawein/outpost
# converge One review pass finds some issues. `converge` runs the whole check set, fixes what it finds, and re-runs until the work is clean or it hits a hard round cap. Clean means zero blockers and zero majors after a full check run; a confirmed minor is reported, not required to be fixed. Use it before something ships when a single check is not enough: a diff, a design, a release. The cap is the safety rule. Without it, the loop never ends or weakens the checks until they pass. `converge` stops at the cap and reports what it could not fix. It never declares clean when it is not. ## When to use it - Before shipping work that needs more than one pass: a diff, a design, or a plan. - When lint, tests, and review have not all been run together on the same work in one go. - In Claude Code only. `converge` ships to Claude alone; the other tools' installs do not carry it, because the loop needs a host that runs checks and fixes on its own. - Skip it for a small, reversible change where a single `code-review` is enough. ## Required inputs - The work to clean: a diff, a design doc, a plan, or a release. - Which checks apply: lint, tests, `code-review`, `grill`, `self-refute`, and optionally `prove` (when the work carries key numbers) or `premortem` (when it is a plan or a launch). ## Steps 1. Run every applicable check on the work. Collect all findings into one list, each with a severity: blocker, major, or minor. Run the deterministic checks first (lint, tests), so a cheap