How do I install the Multiplan skill?

Run npx skillmds@latest add cyperx84/multiplan in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Multiplan safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Multiplan?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Multiplan free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Multiplan?

cyperx84 (@cyperx84) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Multiplan by cyperx84 — Agent Skill

Name: Multiplan
Author: cyperx84

Multiplan

Run multi-model parallel planning workflows using the multiplan CLI. Sends a task simultaneously to Claude (correctness lens), Gemini (scale lens), Codex (speed lens), and GLM-5 (critic lens), cross-examines the results, then converges on a weighted final plan. Use when the user asks to: plan something with multiple models, run a multiplan, get a stress-tested technical plan, use multiplan on a task, plan with all models, or run parallel planning. NOT for simple one-model plans — use this when the user explicitly wants multi-model or multiplan.

cyperx84 8de304c 3.3 KB Updated Sep 17, 2026

File contents

multiplan

Run a task through 4 models in parallel → cross-examine → converge on a weighted final plan.

Binary: multiplan (installed at ~/.local/bin/multiplan) Runs saved to: ~/.multiplan/runs/<timestamp>/

Setup check

multiplan --version   # should print "multiplan v0.4.0"

Required env vars (at least one model must be available):

ANTHROPIC_API_KEY — Claude (Opus)
GOOGLE_AI_API_KEY or GEMINI_API_KEY — Gemini
OPENAI_API_KEY — Codex/GPT
ZAI_API_KEY — GLM-5

Basic usage

# Run planning on a task
multiplan "Design a rate limiting system for the API"

# With requirements and constraints
multiplan "Build a real-time notification system" \
  --req "Must support 10k concurrent users, WebSocket-based" \
  --con "No new infrastructure — use existing Redis + Postgres"

# Verbose (see each model finish in real time)
multiplan "Design an auth system" --verbose

# JSON output (for piping/scripting)
multiplan "Design a caching layer" --json

# Quiet (errors + final plan only)
multiplan "Design a webhook system" --quiet

Output files

After a run, ~/.multiplan/runs/LATEST/ contains:

File	Contents
`plan-claude.md`	Correctness + edge cases lens
`plan-gemini.md`	Scale + ops simplicity lens
`plan-codex.md`	Implementation speed lens
`plan-glm5.md`	Failure analysis + critic lens
`debate.md`	Cross-examination of all 4 plans
`final-plan.md`	✅ Weighted synthesis — start here

Eval

# Eval a single plan
multiplan eval ~/.multiplan/runs/LATEST/final-plan.md

# Eval all plans in a run directory
multiplan eval ~/.multiplan/runs/LATEST/

# With a fixture (task + expected topics + min score)
multiplan eval ~/.multiplan/runs/LATEST/ --fixture eval/fixtures/rate-limiter.json

# JSON output
multiplan eval ~/.multiplan/runs/LATEST/ --json

Config file

Create ~/.config/multiplan/config.yml to set defaults:

debate_model: claude
converge_model: claude
timeout_ms: 120000
verbose: false

CLI flags always override config file.

Workflow for user requests

Extract the task, requirements, and constraints from what the user said
Run multiplan "<task>" --req "<reqs>" --con "<constraints>" --verbose
When complete, read ~/.multiplan/runs/LATEST/final-plan.md and present it
Optionally run multiplan eval ~/.multiplan/runs/LATEST/ and show scores

Tips

Always use --verbose so the user can see models completing in real time
If a model is unavailable (missing API key), multiplan skips it and continues with the rest
The final plan is already a synthesis — no need to re-summarise it; just present it directly
For big tasks, add --timeout 180000 (3 min) to give models more time

---name: multiplandescription: Run multi-model parallel planning workflows using the multiplan CLI. Sends a task simultaneously to Claude (correctness lens), Gemini (scale lens), Codex (speed lens), and GLM-5 (critic lens), cross-examines the results, then converges on a weighted final plan. Use when the user asks to: plan something with multiple models, run a multiplan, get a stress-tested technical plan, use multiplan on a task, plan with all models, or run parallel planning. NOT for simple one-model plans — use this when the user explicitly wants multi-model or multiplan.---# multiplanRun a task through 4 models in parallel → cross-examine → converge on a weighted final plan.**Binary:** `multiplan` (installed at `~/.local/bin/multiplan`)**Runs saved to:** `~/.multiplan/runs/<timestamp>/`## Setup check```bashmultiplan --version # should print "multiplan v0.4.0"```Required env vars (at least one model must be available):- `ANTHROPIC_API_KEY` — Claude (Opus)- `GOOGLE_AI_API_KEY` or `GEMINI_API_KEY` — Gemini- `OPENAI_API_KEY` — Codex/GPT- `ZAI_API_KEY` — GLM-5## Basic usage```bash# Run planning on a taskmultiplan "Design a rate limiting system for the API"# With requirements and constraintsmultiplan "Build a real-time notification system" \ --req "Must support 10k concurrent users, WebSocket-based" \ --con "No new infrastructure — use existing Redis + Postgres"# Verbose (see each model finish in real time)multiplan "Design an auth system" --verbose# JSON output (for piping/scripting)multiplan "Design a caching layer" --json# Quiet (errors + final plan only)multiplan "Design a webhook system" --quiet```## Output filesAfter a run, `~/.multiplan/runs/LATEST/` contains:| File | Contents ||------|----------|| `plan-claude.md` | Correctness + edge cases lens || `plan-gemini.md` | Scale + ops simplicity lens || `plan-codex.md` | Implementation speed lens || `plan-glm5.md` | Failure analysis + critic lens || `debate.md` | Cross-examination of all 4 plans || `final-plan.md` | ✅ Weighted synthesis — start here |## Eval```bash# Eval a single planmultiplan eval ~/.multiplan/runs/LATEST/final-plan.md# Eval all plans in a run directorymultiplan eval ~/.multiplan/runs/LATEST/# With a fixture (task + expected topics + min score)multiplan eval ~/.multiplan/runs/LATEST/ --fixture eval/fixtures/rate-limiter.json# JSON outputmultiplan eval ~/.multiplan/runs/LATEST/ --json```## Config fileCreate `~/.config/multiplan/config.yml` to set defaults:```yamldebate_model: claudeconverge_model: claudetimeout_ms: 120000verbose: false```CLI flags always override config file.## Workflow for user requests1. Extract the task, requirements, and constraints from what the user said2. Run `multiplan "<task>" --req "<reqs>" --con "<constraints>" --verbose`3. When complete, read `~/.multiplan/runs/LATEST/final-plan.md` and present it4. Optionally run `multiplan eval ~/.multiplan/runs/LATEST/` and show scores## Tips- **Always use `--verbose`** so the user can see models completing in real time- If a model is unavailable (missing API key), multiplan skips it and continues with the rest- The final plan is already a synthesis — no need to re-summarise it; just present it directly- For big tasks, add `--timeout 180000` (3 min) to give models more time

Frequently asked questions