do-in-steps — Claude Code skill: run a task step by step
Splits a complex task into dependent steps, runs them in order, and has a separate judge check each step before the next one starts.
npx skills add NeoLabHQ/context-engineering-kit --skill do-in-steps -a claude-codeWhat do-in-steps does
The skill acts as an orchestrator and keeps its own hands off the code. It breaks the task into ordered subtasks, then sends each one to a fresh sub-agent with an isolated context window. A summary of earlier outputs travels with the prompt, so the chain does not drag the whole repository along.
Verification has two layers. A meta-judge runs alongside the implementer and writes criteria for that particular step. Once both finish, an independent judge scores the result against those criteria. A failing step is retried up to three times with the same specification, and the meta-judge is not called again.
Model choice happens per step, not per run: sonnet and haiku by default, opus only when a step earns it under the skill's own rules. The --model flag overrides that for every sub-agent, while --strict removes leniency from scoring. It relies on sub-agents and the Task tool, which Claude provides.
What it helps with
- break a refactor into steps where each depends on the last
- check every step's output before moving on
- carry context between subtasks without flooding the window
- match model size to the difficulty of a single step
- get an independent review of a sub-agent's work
Example prompts for the agent
- Run
do-in-stepsto renameUserServiceand update every consumer - Do this in steps
do-in-steps: migrate the database schema, then fix the queries - Break the API auth work into steps
do-in-stepsand add tests
How to install do-in-steps
1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-in-steps -a claude-code2Manually: put the files in place
~/.claude/skills/do-in-steps/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/do-in-steps/
1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-in-steps -a codex2Manually: put the files in place
.agents/skills/do-in-steps/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-in-steps -a cursor2Manually: put the files in place
.cursor/skills/do-in-steps/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install NeoLabHQ/context-engineering-kit do-in-steps2Manually: put the files in place
.github/skills/do-in-steps/SKILL.md1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-in-steps -a gemini-cli2Platform CLI
gemini skills install https://github.com/NeoLabHQ/context-engineering-kit --path skills/do-in-stepsManually: put the files in place
.devin/skills/do-in-steps/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-in-steps -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about do-in-steps
How do I install do-in-steps in Claude Code?
Run npx skills add NeoLabHQ/context-engineering-kit --skill do-in-steps -a claude-code. The skill lands at ~/.claude/skills/do-in-steps/SKILL.md.
What does do-in-steps do?
It decomposes a complex task into ordered subtasks and gives each one its own sub-agent. A separate judge checks the step against criteria produced by a meta-judge.
How do --strict and --model differ?
--strict drops the leniency rule, so a step passes only on a high judge score and otherwise gets retried. --model forces one model on all sub-agents instead of picking a tier per step.
