Opencode Ultraplan — Development skill for Claude Code
Deep multi-agent planning with adversarial critique.
How to install Opencode Ultraplan
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open rianprei/opencode-ultraplan and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Opencode Ultraplan does
Deep multi-agent planning with adversarial critique. Local replica of Anthropic's dead /ultraplan — built from actual leaked source code. 100% local, no cloud, no quota. Runs on opencode, Claude Code, Codex, Grok.
Alternatives in Development
- 17 Auto Mode Critique — Auto Mode Critique Prompt 2.3k ★
- Compound Learning — Compound Learning & Deep Planning 1.1k ★
- Claudecode — Open Source ClaudeCode Leaked 1k ★
README
opencode-ultraplan
**Local replica of Anthropic's `/ultraplan`** — deep multi-agent planning with adversarial critique, built from the actual leaked source code (npm v2.1.88), the best of 6 open-source replicas, GitHub's spec-kit, and the superpowers methodology. Runs 100% locally. No cloud. No quota. No browser. No limits.
[](LICENSE) [](CHANGELOG.md) [](#installation) [](#the-ultraplan-is-dead) [](#the-ultraplan-is-dead)
The ultraplan is dead
Anthropic shipped `/ultraplan` in Claude Code **v2.1.92** (April 2026) and killed it in **v2.1.222** (July 2026). It was a cloud session where Opus researched for up to 30 minutes with 3 parallel exploration agents + 1 critique agent, producing a plan you reviewed in the browser like a PR.
It's gone. This is the restoration.
What it does
/ultraplan Spawns a **frota of 3 parallel explorers + 1 adversarial critic** that reads your codebase, hunts for reusable code, tries to break the draft plan, and produces a structured plan with:
- Complexity routing (score 4–12 → direct / 3+1 fleet / 3+1 + file-ownership table)
done =observable criteria per step — catches fake-progress at step 2 instead of step 10PRECISA CLARIFICARmarkers — zero silent guessing- Section-by-section revision —
/ultraplan revise :regenerates ONLY section n - 3 execution surfaces — direct / subagent-per-step / remote PR
- Adversarial critic with refutation loop — up to 2 rounds, then it's visible, never deadlock
Architecture
┌────────────────────────────────────────────────────────Related Skills
ClaudeCode Leak
A comprehensive technical autopsy and archival index of the March 31, 2026 Anthropic Claude Code source leak.
Island Mode Server
Agent Skill: build an on-premise island-mode fallback server — live two-way RxDB replica of Firestore, LAN fai
Agy Review
Adversarial critique via agy using Gemini Pro — independent second opinion on code, plans, or arguments
Critique Adversarial
/critique-adversarial — Hostile cut-editor for articles. Reads the full piece first, maps the thesis spine, th
Sdd
Produce a formal specification through adversarial critique.
Mythos Router
The leaked Anthropic reasoning protocol. Running locally. Zero-drift coding with Strict Write Discipline and a
Related Agents
Secret Hunter
Hunts publicly-leaked credentials (API tokens, cloud keys, DB strings) that belong to a previously-ingested pr
Competitor
Role-plays a fictional rival's reaction to the proposal, framed as leaked competitive intelligence — outside-i
Kimi Worker
Forward a coding, diagnosis, or review task to Kimi K3 (Moonshot) through the local kimi-worker.sh runtime. Us
