Claude Skill MartBrowse skillsWhat is a Skill?
 
← Back to list

Ultrawork Mode (Maximum-Precision Work Discipline)

Type 'ultrawork' in your prompt and the agent switches to a maximum-precision mode that demands failing-first proofs, real-surface evidence, and cleanup receipts before claiming done.

Dev & CodingAdvanced★ 67,531⑂ 5,507AI score 7/10Last updated: Aug 9, 2026

What it does

When the keyword ultrawork (or ulw) appears, the agent enters a "CODE RED" precision mode built on one rule: never claim done from inference, only from captured evidence.

  • Tier triage (LIGHT/HEAVY): touching auth, DB migrations, external integrations, concurrency, or cross-domain refactors escalates to HEAVY — 3+ success criteria plus a reviewer loop.
  • Binding goal contract: registers the deliverable, per-criterion PASS/FAIL observables, the evidence artifact, and an explicit "when to stop" line.
  • Durable notepad: an append-only mktemp markdown file holding plan, findings, RED/GREEN captures and artifact paths so state survives context compaction.
  • Execution loop PIN → RED → GREEN → SURFACE → CLEAN: capture the failing proof before production code, make the smallest change to flip it green, re-prove on the real surface (curl -i, real Chrome, a real pty rendered through xterm.js), then tear down servers/tmux/containers/temp files and log a cleanup receipt.
  • Search discipline: LSP for symbols, ast-grep for structural shapes, codegraph for architecture, explorer/librarian subagents for wide discovery and external research.

Who it's for

  • Developers tired of "it's done" claims that don't actually work.
  • Teams that want TDD and manual-QA evidence genuinely enforced.
  • Advanced Codex/omo users orchestrating parallel subagents.
  • Overkill for quick Q&A or throwaway scripts.

Examples

  1. API bug fix: "ulw the login endpoint returns 200 for malformed emails" → capture the RED unit test → minimal fix → curl -i proof of the 400 status/body → atomic commit referencing the evidence path.
  2. TUI visual QA: after a terminal layout change, render through the real xterm.js web terminal and screenshot it instead of tmux capture-pane, then record the tmux kill-session cleanup receipt.
  3. New payment webhook (HEAVY): register four criteria (happy path, boundary payload, duplicate event, adjacent-surface regression) and re-capture delta evidence until the reviewer subagent approves unconditionally.

· · · Install guide · · ·

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the packages/omo-codex/plugin/components/ultrawork/skills/ultrawork folder from the GitHub repo code-yeongyu/oh-my-openagent into my ~/.claude/skills/ultrawork/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/code-yeongyu/oh-my-openagent.git && mkdir -p ~/.claude/skills && cp -r oh-my-openagent/packages/omo-codex/plugin/components/ultrawork/skills/ultrawork ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal and ensure the skills folder exists: mkdir -p ~/.claude/skills
  2. Clone the repository: git clone https://github.com/code-yeongyu/oh-my-openagent.git
  3. Copy just the skill directory: cp -r oh-my-openagent/packages/omo-codex/plugin/components/ultrawork/skills/ultrawork ~/.claude/skills/
  4. Verify that ~/.claude/skills/ultrawork/SKILL.md exists.
  5. Restart Claude Code (or Codex) and confirm ultrawork shows up in the skill list.
  6. Trigger it by prefixing your request with ultrawork or ulw, e.g. ulw fix the 400-response bug in the payments API.
  7. (Optional) Upstream, an omo hook injects this file automatically. Without that hook, edit tool names such as create_goal, update_plan, and multi_agent_v1.* to match your environment's TODO/subagent equivalents.
View source on GitHub ↗License: NOASSERTION