AI-Native Product Team Skills
Reusable Claude Code skills for the AI-Native Product Team course. Installable as a Claude Code plugin marketplace.
Available Skills
/team-assess - Team AI Autonomy Assessment
Interviews you about your product team's AI maturity across 6 dimensions, generates a structured assessment report and a visual radar chart.
What it does:
- Walks through 6 dimensions (Research, Specs, Prototyping, Analytics, Quality, Team Workflows)
- Provides calibration examples at each level using the self-driving car metaphor
- Asks for current score (0-5) and 6-month target for each dimension
- Generates
team-assessment.md with ratings, observations, and priority recommendations
- Generates
radar-chart.svg comparing current vs target scores
/council - Multi-Perspective AI Council
Get 3-5 expert viewpoints on any product question, with adversarial debate and synthesis. Use for strategic decisions, competitive analysis, feature prioritization, or any question where multiple perspectives improve the outcome.
What it does:
- Assembles a panel of 3-5 expert personas tailored to your question type
- Each persona evaluates independently via sub-agents (no groupthink)
- Highlights agreements, disagreements, and the key tension in the debate
- Produces a synthesis with clear recommendations and confidence levels
- Saves output as
council-[topic-slug].md
Example invocations:
/council Should we build feature X or feature Y for Q2?/council What's the best go-to-market strategy for our new product?/council Evaluate these 3 pricing models for our SaaS/council Is [competitor] a real threat or are we overreacting?
/prototype - Autonomous Prototype Builder
Build any kind of working interactive prototype, then autonomously improve it using the Generator-Evaluator loop. Based on Anthropic's harness architecture (March 2026).
What it does:
- Calibrates evaluation criteria based on prototype type (dashboard, flow, page, tool - suggests domain-specific criteria, you confirm)
- Plans a prototype spec from available context (CLAUDE.md, competitive analysis, brand guidelines - works with minimal context too)
- Spawns a Generator sub-agent to build a multi-file interactive prototype
- Spawns an Evaluator sub-agent to independently critique it with browser automation
- Loops: Generator improves based on Evaluator feedback, Evaluator re-scores
- Runs 3-5 rounds autonomously until quality threshold is met
- Produces: working prototype, improvement-log.md, product-passport.md
Three-agent architecture (Orchestrator + Generator + Evaluator) ensures the Evaluator is genuinely independent - separate context window, no access to the Generator's reasoning.
Example invocations:
/prototype Landing page for our B2B sales tool/prototype Admin dashboard showing real-time pipeline metrics/prototype Onboarding flow for first-time users/prototype Settings page with team management and billing/prototype Data visualization of monthly revenue trends
/autoresearch - Autoresearch Loop for Any Work
Apply Karpathy's autoresearch loop (goal + mechanical fitness + mutable surface + keep-or-revert iteration) to ANY measurable workflow - code, content, sales, research, design, operations. Not just ML or software. Includes a pre-loop triage that refuses fat-tailed, reflexive, or slow-feedback problems without adapting the mode.
What it does:
- Stage 1: Intake - captures goal, candidate metric, mutable surface, guard metrics, budget
- Stage 2: Triage - scores the candidate across 5 dimensions (feedback latency, metric mechanicality, tail shape, sample size, surface locality) and refuses ill-shaped problems with alternatives
- Stage 3: Fitness design - enforces 6 hard requirements (callable command, fixed corpus, baseline, significance threshold, sanity check, guard metrics)
- Stage 4: Mode selection - picks one of five modes based on domain
- Stage 5: Loop + post-mortem - scaffolds
goal.md, runs the keep-or-revert loop, diagnoses plateau causes if reached
Five modes for different domains:
- Pure optimizer - thin-tailed, fast-feedback, clean metric (code, ML, compression, SQL tuning)
- Barbell - 85% proven baseline + 15% wild experiments (fat-tailed, reflexive: B2B sales, content, design)
- Via negativa - "what to kill" not "what to add" (fragile, over-engineered systems)
- Inverted - optimize rejection of bad, not selection of good (imbalanced classes like lead qualification)
- Human-in-loop - agent proposes, human decides (slow feedback, irreversible actions)
haabe