11 Claude Code skills that catch bad bets before you ship, sharpen every decision, and compound your productivity and ROI over time.
Maintained by Varun Kulkarni · Setup ↓ · Skills ↓ · Decision Sequence ↓
A set of Claude Code slash commands organized around how product decisions actually flow. Each skill targets a specific decision failure mode. A CLAUDE.md file holds your product context, decision history, and known biases so every skill has the full picture.
No app. No API. Markdown files in a folder.
| # | Command | What It Does |
|---|---|---|
| 1 | assumption-check |
Surface 5 ranked untested assumptions with cheapest validation tests |
| 2 | downside-case |
Build the strongest possible case for NOT building a feature |
| 3 | 10x-or-10-percent |
Evaluate whether a bet is incremental improvement or transformative change |
| # | Command | What It Does |
|---|---|---|
| 4 | scope-creep-detector |
Flag every scope-expanding sentence in a PRD or spec |
| 5 | pre-mortem |
Write a realistic failure post-mortem from 6 months in the future |
| 6 | strategy-smell-test |
Apply 7 smell tests to detect weak strategy disguised as good strategy |
| # | Command | What It Does |
|---|---|---|
| 7 | stakeholder-translator |
Show how 4 stakeholders will actually read your message |
| 8 | say-no-script |
Generate 3 pushback scripts for stakeholder requests you need to decline |
| 9 | exec-summary-sharpener |
Find weak spots in executive-facing documents without rewriting them |
| # | Command | What It Does |
|---|---|---|
| 10 | decision-audit |
Analyze your decision journal to reveal patterns, biases, and improvement areas |
| 11 | portfolio-validation |
Grade your product instincts by analyzing bets you missed or got wrong |
The LEARN layer is what changed things for me. Log your decisions in the journal, run decision-audit (with the /user: or /project: prefix that matches your install), and get a report showing where you optimize for speed when experimentation wins, where you skip structured evaluation, and where your confidence doesn't match your outcomes.
- Claude Code installed and authenticated
git clone https://github.com/varunk130/AI-Builder-Decision-Analyst.git
cd AI-Builder-Decision-Analyst
# Install the skills into your user-level Claude Code commands directory
mkdir -p ~/.claude/commands
cp skills/*.md ~/.claude/commands/
# Then run any command from any project
claude "/user:assumption-check We're building a self-serve analytics dashboard for SMB customers"mkdir -p your-project/.claude/commands
cp AI-Builder-Decision-Analyst/skills/*.md your-project/.claude/commands/
cp -r AI-Builder-Decision-Analyst/templates/ your-project/templates/mkdir -p ~/.claude/commands
cp AI-Builder-Decision-Analyst/skills/*.md ~/.claude/commands/
# Now use /user:assumption-check from any project📝 Slash-command prefix: Claude Code namespaces commands by install location. Use
/user:<command>when installed at user level (Options 1 and 3) and/project:<command>when installed at project level (Option 2). The command tables below show the bare command name — pick the prefix that matches your install.
- Open
templates/decision-journal.md - Log each decision using the template format (takes ~2 minutes per entry)
- After 15-20 entries, run
decision-audit - Review your Decision Intelligence Report
- Run again in 90 days to track how your patterns shift
The journal captures: what you decided, what you rejected, what you optimized for, who influenced you, your confidence level, and the outcome.
For major product decisions, run these in order:
Use the prefix that matches your install (/user: for Options 1 and 3, /project: for Option 2):
1. assumption-check → Find what you don't know
2. downside-case → Stress-test the idea
3. 10x-or-10-percent → Clarify the bet size
4. scope-creep-detector → Tighten the spec
5. pre-mortem → Anticipate failure
6. strategy-smell-test → Validate the strategy
If your idea survives all 6, ship it with confidence.
Be specific in your input. "We're building a dashboard" is weak. "We're building a self-serve analytics dashboard for SMB e-commerce brands who currently export CSV reports weekly and share them in Slack" is strong.
Disagree with the output. These skills are designed to push back. If you can refute every argument, your thinking is solid. If you can't refute one — you found a blind spot.
Log everything. The LEARN layer gets sharper the more decisions you log. 20 entries is the minimum for meaningful pattern detection.
AI-Builder-Decision-Analyst/
├── README.md
├── skills/
│ ├── assumption-check.md # DECIDE layer
│ ├── downside-case.md
│ ├── 10x-or-10-percent.md
│ ├── scope-creep-detector.md # BUILD layer
│ ├── pre-mortem.md
│ ├── strategy-smell-test.md
│ ├── stakeholder-translator.md # COMMUNICATE layer
│ ├── say-no-script.md
│ ├── exec-summary-sharpener.md
│ ├── decision-audit.md # LEARN layer
│ └── portfolio-validation.md
├── templates/
│ ├── decision-journal.md # Log your decisions here
│ └── anti-portfolio.md # Log your misses here
└── assets/ # Carousel images
See CONTRIBUTING.md for scope, branch / PR flow, and the skill file format.
Part of a portfolio of AI agent and skill libraries for product, GTM, and decision-making teams.
Discovery & research
- ai-customer-discovery-skills - Turn raw customer signal into validated product opportunities (12 skills planned)
- jtbd-extractor - Extract Jobs-to-be-Done statements from research, with opportunity scoring
Strategy & decisions
- claude-code-skills - 29 production-grade skills for finance, product, strategy, and game theory
Go-to-market
- ai-gtm-skill-library - 31 opinionated GTM skills across the full discover -> renew lifecycle
- ai-marketing-claude-skills - 12 marketing-ops skills with scoring algorithms and statistical frameworks
- ai-partner-ecosystem-analysis - Deep research on any ISV, partner, or competitor with a 1-slide PPTX output
UX & design
- ai-ux-skill-library - 12 frameworks for designing UX for AI products, agents, and AI-powered experiences
Multi-agent demos
- ai-pm-agents-suite - 6-agent pipeline plus 3 standalone PM agents (decision engine, financial analyst, stakeholder translator) that turn customer feedback into strategy, PRDs, and comms
- ai-legal-team-agent - 4-agent legal analysis team with Python orchestrator and Claude Code skills
Evaluation & operations
- AI-Eval-Skills - 6 skills to plan, generate, run, interpret, and triage AI agent evaluations
- ai-workflow-playbooks - 21 playbooks + 10 skills + 4 guardians + 5 runbooks across the 7-stage delivery pipeline
MIT — see LICENSE for the full text.
All decision data in the templates is fictional and anonymized. The entries are examples to show how the journal and audit work. Replace them with your own decisions to get real value from the LEARN layer.
Built by Varun Kulkarni — AI Product Manager building tools that help AI builders 10x their impact. Star the repo or leave feedback if it's useful.