PM toolkit for Claude Code. 9 skills that cover the full product management lifecycle: spec writing, doc reviews, prioritization, impact sizing, competitive analysis, and launch planning.
claude plugin add prodkit| Skill | Command | What it does |
|---|---|---|
| Feature Spec Interview | /feature-spec-interview |
Structured interview that produces gap-free feature specs. 14 prompts, 61 question groups, NLSpec format (WHAT/WHEN/WHY/VERIFY). |
| Feature Doc Review Panel | /feature-doc-review-panel |
Multi-agent review from 7 perspectives: Engineering, Design, Executive, Legal, UX Research, Skeptic, Customer Voice. |
| Decision Doc | /decision-doc |
Document strategic decisions with rationale, alternatives, and trade-offs. |
| Skill | Command | What it does |
|---|---|---|
| Prioritize | /prioritize |
Classify tasks using the LNO Framework (Leverage/Neutral/Overhead). Target: 40% L, 35% N, 20% O. |
| Impact Sizing | /impact-sizing |
Quantify feature value with driver trees, confidence levels, and a 4-step sizing framework. |
| Skill | Command | What it does |
|---|---|---|
| Define North Star | /define-north-star |
Identify and validate your North Star Metric. Covers formula design, input metrics, and guardrails. |
| Competitor Analysis | /competitor-analysis |
Deep competitive research or ongoing monthly monitoring. Checks internal context first, fills gaps with web search. |
| User Research Synthesis | /user-research-synthesis |
Turn raw interview notes and transcripts into actionable product insights with recurring themes. |
| Skill | Command | What it does |
|---|---|---|
| Launch Checklist | /launch-checklist |
Generate prioritized launch checklists with owners, dependencies, and critical path identification. |
Each skill is a structured prompt that guides Claude Code through a specific PM workflow. Skills ask clarifying questions, check your project context, and produce artifacts (docs, analyses, checklists) saved to your workspace.
Skills save outputs to two locations: projects/<name>/ for feature-scoped work, and knowledge/wiki/ for cross-cutting knowledge. Raw source documents go in knowledge/raw/.
Prodkit uses a three-layer knowledge system inspired by Karpathy's LLM Wiki pattern. Instead of re-deriving context from scratch every session, knowledge compounds over time in a persistent wiki that the LLM maintains for you.
knowledge/
raw/ <- Your original documents. LLM reads, never writes.
wiki/ <- LLM-maintained pages. Entities, concepts, syntheses, decisions.
INDEX.md <- Front page. LLM reads this first to navigate.
projects/
<feature-name>/ <- Feature specs, review reports, launch checklists.
These folders ship with the plugin. Browse the READMEs inside each one to see how they work.
Three layers:
-
Raw sources (
knowledge/raw/). Drop your documents here: interview transcripts, strategy decks, analytics exports, competitor screenshots. Prodkit reads them but never modifies them. They're the source of truth. -
The wiki (
knowledge/wiki/). Prodkit writes and maintains this. Entity pages for people and companies. Feature pages that aggregate specs, sizing, and decisions. Concept pages for frameworks. Comparison pages for competitors. Cross-linked, backlinked, and kept consistent as new sources arrive. Start with the INDEX.md. -
The schema (
CLAUDE.md). Tells the LLM how to ingest sources, query the wiki, and maintain it over time. You and the LLM co-evolve this as you figure out what works for your domain.
Three operations:
- Ingest. Drop a source into
raw/, tell prodkit to process it. It extracts entities, updates wiki pages, flags contradictions with existing knowledge, and logs the operation. - Query. Ask a question. Prodkit reads the wiki index, loads relevant pages, and synthesizes an answer. Good answers get filed back as new wiki pages so your explorations compound.
- Maintain. Periodically, prodkit suggests cleanup: find contradictions, flag stale pages, add missing cross-links. You approve before anything changes.
Skills feed the wiki automatically. When you run /feature-spec-interview, prodkit creates a feature page. /competitor-analysis updates entity pages. /impact-sizing adds data to the feature page. /decision-doc creates a decision page with backlinks. Over time, the wiki becomes a comprehensive, cross-referenced product knowledge base that makes every future skill run better informed.
The wiki is optional. Skills work without it. But once it starts building, each session gets smarter than the last.
The headliner skill. It runs a structured interview with 14 prompts and 61 question groups to produce behavioral contracts detailed enough that a new engineer (or AI agent) can execute with at most one clarifying question.
Key concepts:
- NLSpec format: Every behavior has WHAT, WHEN, WHY, and VERIFY components
- The New-Hire Test: A spec is complete when a capable new hire with no context could implement it
- Skip Matrix: Not every question applies. The interview adapts based on your feature's characteristics
- Two-tier output: Tier 1 (behavioral contract for execution) + Tier 2 (strategic context for humans)
- Gate check: Before any interview, 4 gate questions validate the feature should enter the pipeline at all
Prodkit includes 7 passive automation systems that make skills smarter across sessions. No setup required. They activate automatically via CLAUDE.md.
Ask natural language questions and prodkit routes them to the right data source. Analytics MCPs for metrics queries, PM tools for ticket status, workspace files as fallback. If no MCP is connected, it works with local files silently.
Tracks which skills you use, what corrections you make, and how accurate your estimates are. Stored in knowledge/prodkit-learning-log.md (created on first event, not at install). After 20+ entries, suggests a monthly review.
Two destinations: projects/<name>/ for feature-scoped work (specs, reviews, checklists), knowledge/wiki/ for everything else (research, competitor analysis, decisions, sizing, strategy). The test: can you name the feature? If yes, projects/. If not, knowledge/wiki/.
If you run the same skill twice in one day, prodkit asks whether to update the existing file or create a new version. Every output gets a hidden version header for tracking. No silent overwrites.
After 3+ similar corrections (e.g., "too long" three times), suggests updating your style preferences. After 3+ repeated skill sequences, suggests combining them. Flags stale context files older than 2 weeks. Notes missing upstream data that would improve output quality.
When processing 3+ independent items (multiple interviews, competitors, or review perspectives), prodkit spawns parallel sub-agents instead of working sequentially. Synthesizes results into a single output and flags conflicts.
Skills reference each other's outputs automatically. /feature-spec-interview output feeds into /feature-doc-review-panel. /impact-sizing estimates get calibrated against actual post-launch results. /user-research-synthesis outputs enrich future spec interviews. The full dependency map is in CLAUDE.md.
- Claude Code CLI or Desktop
MIT