A 7-stage Claude Code skill pipeline that puts any decision, claim, or strategy through rigorous deliberation — combining devil's advocacy, community research, fact verification, operational analysis, adversarial debate, judgment, and a phased action plan.
| # | Stage | Skill | What it does |
|---|---|---|---|
| 1 | Devil's Advocate | devils-advocate |
Challenges premises, exposes hidden assumptions, surfaces hard constraints |
| 2 | Community Scan | last30days |
Practitioner stories from Reddit/X/Web (last 30 days) |
| 3 | Research & Verify | research |
Perplexity-powered fact-checking of community claims |
| 4 | COO Systems Analysis | coo |
Operational lens with explicit revenue math |
| 5 | Adversarial Debate | (Opus agents) | 2–3 agents, 3 rounds, with concessions |
| 6 | Judge | judge |
5-criterion scorecard, picks a winner |
| 7 | Action Plan | action-plan |
Week/month/quarter/year plan with checkpoints + flip conditions |
Clone into your Claude Code skills directory:
cd ~/.claude/skills
git clone https://github.com/Mattyreed1/deep-deliberation.git temp-dd
mv temp-dd/* .
rm -rf temp-ddOr symlink individual skills as needed.
Trigger the full pipeline:
/deep-deliberation should I switch from Stripe to Lemon Squeezy?
Or invoke any sub-skill standalone:
/devils-advocate <claim>
/last30days <topic>
/research <question>
/coo <operational problem>
/judge <debate transcript>
/action-plan <verdict>
- Claude Code (CLI or Desktop)
- For
last30days: optionalOPENAI_API_KEY(Reddit) andXAI_API_KEY(X/Twitter) in~/.config/last30days/.env - For
research: Perplexity MCP server configured
- Community before research — surfaces specific numeric claims that research then verifies
- COO after both research stages — operator's lens needs practitioner reality + verified facts
- Debate after options are grounded in math — philosophy without numbers is noise
- Judge before action plan — you need a committed winner before you can operationalize
MIT