nraford7

Do It Skill — Testing skill for Claude Code

Testing community

Claude Code skill: single-shot autonomous build pipeline (spec → plan → build → review → commit) with tiered review and a diminishing-returns judge.

How to install Do It Skill

This entry records only its repository, not the path inside it, so there is no exact command to give. Open nraford7/do-it-skill and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Do It Skill does

Claude Code skill: single-shot autonomous build pipeline (spec → plan → build → review → commit) with tiered review and a diminishing-returns judge.

Alternatives in Testing

OpenAgentsControl — AI agent framework for plan-first development workflows with approval-based execution 4.8k ★
Factory — Dark Factory Mode - Spec-in, software-out autonomous pipeline 2.8k ★
Production Grade — Production Grade — route any build, feature, review, test, harden, ship, or docs request through the 14-agent 175 ★

README

do-it — Autonomous Build Pipeline Skill

A [Claude Code](https://claude.com/claude-code) skill for single-shot, end-to-end builds. You give one instruction; the pipeline drives spec → plan → build → review → commit → push without asking permission at every gate.

What it does

[0] Clarify (only if blocking)
[1] Spec draft   → review loop (tiered: LIGHT / STANDARD / HEAVY)
[2] Plan         → review loop
[3] Execution    (no permission asks)
[4] Post-build review
[5] Verification gate → commit + push

Key ideas:

Review-tier evaluator — a rubric scores each task and picks how heavy the review stack should be (FAST mode skips the loops entirely for low-risk work).
Diminishing-returns judge — an independent judge decides CONTINUE / STOP / SPLIT after each review pass, so review effort stops when it stops paying.
Mandatory artifacts — every run leaves a spec, a plan, and a run manifest in git. No rationalized skips.
Watchdog — scripts/watchdog.sh supervises long external review runs (stall detection, hard ceiling), tunable via FE_* env vars.

Path	Purpose
`do-it/SKILL.md`	The skill itself — pipeline, modes, artifact rules
`do-it/references/evaluator-rubric.md`	Review-tier scoring rubric
`do-it/references/judge-prompt.md`	Diminishing-returns judge prompt
`do-it/scripts/watchdog.sh`	Supervisor for long-running review subprocesses

Install

Copy the `do-it/` directory into your Claude Code skills folder:

cp -R do-it ~/.claude/skills/do-it

Then trigger it with `/do-it ` (or "just build it", "do it end-to-end").

The skill references companion tooling from my setup ([superpowers](https://github.com/obra/superpowers) skills, a `fresheyes` second-model reviewer, and an [Agency](https://github.com/agentbureau/agency) execution layer). It degrades gracefully without them, but the review loops assume an independent reviewer is available.