clawsewitz

The open source AI strategy agent. Now available as a standalone CLI and a Claude Code plugin.

         ███                                                           ███   ███
         ███                                                                 ███
  █████  ███   ███████   ███      ███  ███████    █████  ███      ███  ███  █████  ███████
███      ███  ███   ███  ███      ███  ███      ███      ███      ███  ███   ███      ███
███      ███  █████████  ███ ████ ███   █████   ███████  ███ ████ ███  ███   ███     ███
███      ███  ███   ███   ███ ██ ███       ███  ███       ███ ██ ███   ███   ███    ███
  █████  ███  ███   ███    ███  ███    ███████    █████    ███  ███    ███    ███  ███████

Install

npm (primary):

npm install -g clawsewitz

curl (standalone):

curl -fsSL https://robertsinke.github.io/clawsewitz/install | bash

Claude Code plugin:

curl -fsSL https://robertsinke.github.io/clawsewitz/install-plugin | bash

Skills only:

curl -fsSL https://robertsinke.github.io/clawsewitz/install-skills | bash

Requires: node (>=20.19.0).

Usage

clawsewitz                                         # Interactive strategy session
clawsewitz decompose "Why are customers churning?" # MECE tree with validation
clawsewitz evaluate "Build vs. buy vs. partner"    # Weighted scoring + red team
clawsewitz brief path/to/findings.md               # Answer-first executive brief
clawsewitz plan path/to/recommendation.md          # Roadmap + premortem
clawsewitz audit strategy.md                       # Red-team any plan
clawsewitz research "competitor landscape"         # Competitive intel
clawsewitz doctor                                  # Check installation

What it does

4 agents + 8 workflows + 9 skills. Three layers, composed as the situation demands. No orchestrator walking a fixed sequence.

• Agents — dispatchable specialists (researcher, analyst, writer, challenger)
• Workflows — top-level commands, each a full procedure runnable standalone; the source of truth for how a job gets done
• Skills — discoverable capabilities with natural-language triggers; the agent reaches for the right one when your intent matches

In practice: say "help me stress-test this strategy" → agent triggers red-teaming-plans skill → runs /audit workflow → dispatches challenger agent. You can reach any layer directly.

Workflows

Command	What it does
/cw-decompose <problem>	Break a problem into a validated MECE tree
/cw-evaluate <question>	Score options against criteria and recommend one
/cw-brief <topic>	Write an executive brief or recommendation memo
/cw-plan <recommendation>	Build an implementation roadmap with risk register
/cw-audit <path>	Red-team and premortem an existing plan
/cw-research <query>	Gather competitive, market, or industry intelligence
/cw-case list	List, resume, or archive case workspaces
/cw-frameworks browse	Browse the 70-framework catalog

Skills

Skill	Triggered by
engagement-intake	A fresh strategic brief that needs triaging before analysis
mece-decomposition	"Break this apart", "tree this out", "split the problem"
evaluating-options	"Help me decide", "which should we pick", "weigh these"
writing-briefs	"Draft a memo", "structure this for the board", "one-pager"
implementation-planning	"How do we execute", "90-day plan", "roadmap with owners"
red-teaming-plans	"Poke holes in this", "what could go wrong", "stress-test"
strategic-research	"Competitive landscape", "market sizing", "benchmarks"
framework-library	Look up any of 70 frameworks by name
session-search	Recall prior engagements across cases

Agents

Agent	Role
Researcher	Hunts for evidence across the web, benchmarks, and competitive landscape
Analyst	Decomposes problems, applies frameworks, validates MECE integrity
Writer	Structures findings into briefs, memos, and slide narratives
Challenger	Stress-tests plans, plays adversary, writes the failure narrative

What's under the hood

• 4 agents — Researcher, Analyst, Writer, Challenger
• 8 workflows — decompose, evaluate, brief, plan, audit, research, case management, framework browsing
• 9 skills — 7 capability wrappers over the workflows + 2 utility skills (framework library, session search)
• Enforcement guardrails — auto-commit, analyst amplification on Fail verdicts, session resume, MECE structural check tool
• 70 frameworks — from the Strategist Toolkit: SCQ, HTDQ, Minto Pyramid, Driver Tree, SPADE, Bezos, GTM Stack, Zero-to-One, and 60+ more

Dual Distribution

clawsewitz ships as two distributions:

• Standalone Pi CLI (primary) — install via npm or the standalone curl script. Runs clawsewitz as a top-level command in any terminal, powered by the Pi runtime.
• Claude Code plugin (secondary) — install via the plugin curl script. Adds /cw-* commands and the skill library inside Claude Code sessions.

Both distributions share the same agents, workflows, and frameworks. The CLI is the recommended entry point for new users; the plugin is available for teams already working inside Claude Code.

Version

v1.0.0

License

MIT