Skip to content

Home

Jump to bottom
Nick edited this page Jun 21, 2026 · 15 revisions

Cool Workflow Wiki

Get a saved, cited report from your AI agent — not a chat message you lose. The agent signs its findings (ed25519); anyone can verify them offline with only the public key. CW delegates the model work and keeps the books — it never runs a model itself.

Your AI report, already proven — cw demo tamper forges a signed report three ways and catches every one offline with only the public key

Quick Start

npm install -g cool-workflow
cw -q "What are the main risks here?"

One command. CW auto-detects your repo and agent. Live streaming output included.

Pick an agent:

cw -q "..." -claude      # Claude
cw -q "..." -codex       # Codex
cw -q "..." -deepseek    # DeepSeek

Review any repo — local folder or a URL (no cd needed):

cw -q "What are the risks?" -dir /path/to/project              # a local folder, from anywhere
cw -q "What are the risks?" --link https://github.com/owner/repo   # clone + review a remote repo

--link works with any git host (GitHub/GitLab/Bitbucket/ssh) and with downloadable archives (.tar.gz/.zip). The report records Source: <url>@<commit> and a tamper-evident audit event; cw clones list/gc manage the cached checkouts.

Commands

Command Does
cw -q "question" [-claude|-codex|-deepseek] Full auto architecture review
cw -q "…" -dir PATH Review a local folder from any directory
cw -q "…" --link <url> Review a remote repo by URL (git host or archive)
cw clones list / cw clones gc Inspect / reclaim the cached remote checkouts
cw version Show version
cw update Update to latest release
cw doctor Check setup
cw fix Show fix commands
cw demo tamper Prove the tamper checks in 30s (no agent needed)
cw -q "…" --bundle Seal the run into one portable, verifiable file
cw report verify-bundle <file> Check a report offline — signed findings present + unaltered
cw run <app> --drive --incremental Re-run fast — reuse every unchanged step

How It Works

CW is a TypeScript control-plane with zero runtime deps. It delegates work to YOUR AI agent — it never runs a model itself. Every step is saved, hash-chained, and tamper-evident, and the agent's signed findings can be re-verified offline with only the public key.

ask simple -> run simple -> verify simple -> resume simple

Learn More

Organized from local Obsidian notes and reconciled with the current coo1white/cool-workflow repository state.

Cool Workflow

Start here

Go deeper

Source docs

Clone this wiki locally