Product management
moved to the codebase.

Your PRDs, decks, and strategy docs don’t ship. The codebase does. Acumen puts product craft where it actually matters.

1 Core Skill

15 Commands

7 Reference Docs

Get started Why Acumen?

Works with

01 — The Problem

Product craft happens too far from the codebase.

Decisions are made in docs, slides, and chat — disconnected from what's actually shipped. Without a source of truth, AI product work defaults to generic slop:

— Context evaporates in transit — intent gets paraphrased three times before it reaches the code
— LLMs have no anchor — same vague PRDs, demographic personas, and hedge documents for every product
— Craft loses its feedback loop — no tight cycle between what you decided and what you actually built

Acumen fixes this by embedding product craft where the building happens — commands that read your codebase and an organised knowledge layer before they write a single line.

02 — Commands

16 commands. Each one reads your product's context before it writes a single line.

Each command is a focused product operation — grounded in your codebase and knowledge layer, not generic prompting.

Context

teach

scout

persona

features

Audit

orientation

defensibility

Ideate

diagnose

measure

workshop

Craft

roadmap

increment

critique

Communicate

narrate

changelog

/teach

Context

One-time setup that gathers product context for your project and saves it to .acumen.md. It guides you through exploring the codebase, asking product-focused questions about users, market, business model, strategy, and metrics, then synthesizes findings into persistent context.

Examples

/teach /teach after the pivot to B2B

03 — Context System

Context that lives where the code lives.

Acumen closes the distance with persistent context files that live in your repo, versioned alongside code. Every command reads them before it writes. The more you use Acumen, the sharper the output gets.

.acumen.md

Product context: users, strategy, positioning, constraints, north star metric.

Created by /teach-acumen

.acumen/value-chain.md

End-to-end workflow map per persona — extension points, friction, expansion opportunities.

Created by /teach-acumen · Updated by /workshop

.acumen/personas.md

Behavioral personas grounded in real usage patterns, not demographics.

Maintained by /persona

.acumen/competitors.md

Living competitor map: who they are, what they ship, where they're strong and weak.

Maintained by /scout

.acumen/features.md

Feature inventory: what exists, what's healthy, what's decaying, what's missing.

Maintained by /features

.acumen/value.md

Per-persona value delivery mapped to usage and north star metrics.

Maintained by /value

.acumen/sources.md

Data source config: analytics, database, backlog — how skills pull real data.

Created by /teach-acumen

.acumen/reports/

Point-in-time outputs from diagnostic and ideation skills.

Generated by /diagnose, /measure, /workshop

Value Chain — the backbone

The value chain maps each persona's full job end-to-end. It answers: where do we bring value, where is the user on their own, and where are the expansion opportunities? Eight skills read it. Three maintain it.

Step	Our role	Skills that use it
Identify the problem	None	/diagnose, /workshop
Choose a tool	Assist	/orientation, /scout
Set up & configure	Own	/increment, /measure
Do the core job	Own	/increment, /defensibility
Evaluate results	Assist	/measure, /diagnose
Share & collaborate	None	/workshop, /roadmap

Example value chain. Yours is generated from your product's actual workflow by /teach-acumen.

04 — Anti-Patterns

What distance produces: 12 named tells of AI product slop.

Without codebase context, AI defaults to these patterns. Acumen names each one, teaches the AI to detect them, and refuses to produce them.

Template Filler

Every section filled, none grounded in what’s actually built

Demographic Persona

Demographics because real behavior is observed, not assumed

Inverted Solution

Problem is the solution described negatively — the real friction was never observed

Vanity Metric

Measures things that always go up, not what the product actually tracks

Feature Checklist

Flat list with no awareness of what already exists or how it connects

Generic Risk

“Technical complexity” — because the actual system was never examined

Tautology Story

User story restates the feature because the real job-to-be-done is unknown

Phantom Feature

Proposes what already exists — the codebase was never read

Strategy of Everything

Lists every direction because without context, nothing can be ruled out

Feature Roadmap

Timeline of features with no thesis about what the product is becoming

Greenfield Spec

Scopes from scratch, ignoring existing architecture and shipped features

Consensus Document

Inoffensive to all because without product evidence, there’s nothing to argue from

05 — Get Started

Install in under a minute.

Acumen follows the Agent Skills spec. Drop the skills into your AI tool and go.

Claude Code (project)

# Copy skills into your project
cp -r .claude your-project/
cp -r .agents your-project/

Skills load automatically via Claude Code's skill system.

Claude Code (global)

# Install skills globally
mkdir -p ~/.claude/skills
cp -r .agents/skills/* ~/.claude/skills/

Available in every project on your machine.

Other AI Tools

# Copy the Agent Skills directory
cp -r .agents/skills/ ~/your-tool/skills/

Any tool supporting the Agent Skills spec.

Then run setup

# One-time product context
/teach-acumen

# Start using commands
/diagnose
/orientation
/measure

Works with: Claude Code · Cursor · Windsurf · Gemini CLI · Codex CLI · GitHub Copilot · Kiro · OpenCode · Amp · Cline · Aider · Roo Code · Continue · Zed · Trae · and any tool supporting the Agent Skills spec.

06 — Principles

Why the codebase comes first.

Read before write

Every skill reads the codebase and context layer before producing a single line.
Opinion > comprehensiveness

"RICE doesn't ship products, judgment does."
Brevity > coverage

50-100 lines per skill. Depth lives in reference docs.
Context layer feeds all other layers

Not passive storage — active inputs that shape every artifact.
Context over prompting

Ground every artifact in what actually exists, not what sounds plausible.
Natural flow between skills

Skills suggest the next skill: /diagnose → /workshop → /increment → /critique-product.

07 — What's New & FAQ

What's New

April 9, 2026

Periodic-table skill cards with interactive carousel and embedded skill videos
"Works with" provider icons showing all supported AI tools
9 new skills: /orientation, /defensibility, /measure, /workshop, /narrate, /changelog, /cheatsheet, /value, /feedback
Existing skills rewritten with anti-slop guards, cross-skill flow, and value chain context
/increment added — scope a product increment end-to-end
Unified Skills & Commands section with knowledge-layer breakdown
Copy-to-clipboard on all install code blocks

April 8, 2026

Initial release: 1 core skill, 16 commands, 12 named anti-patterns
Product-thinking skill with 7 reference documents across problem discovery, segmentation, prioritization, metrics, spec craft, stakeholder alignment, and competitive intelligence
Landing page and full documentation

FAQ

How do I install Acumen?

Run npx skills add contenty/acumen for auto-detection, or copy the .agents and .claude directories into your project root.

How do I update?

Run /update-acumen from within your AI tool, or run npx skills update. Your .acumen.md and context files are preserved.

Commands aren't showing up.

Verify the skill files are in the right directory for your tool. For Claude Code, check that .claude/skills/ contains the symlinks. For Cursor, ensure Agent Skills is enabled.

Which tools are supported?

Claude Code, Cursor, Windsurf, Gemini CLI, Codex CLI, GitHub Copilot, Kiro, OpenCode, Amp, Cline, Aider, Roo Code, Continue, Zed, Trae, and any tool supporting the Agent Skills spec.

Is Acumen free?

Everything is Apache 2.0: skills, commands, reference docs — all free and open source.

Do I need /teach-acumen in every project?

Yes, once per project. It generates .acumen.md with your product's context. Without it, commands still work but produce less targeted output.