Skip to content

VGrss/Acumen

Repository files navigation

Acumen

Product fluency for AI. 1 skill, 16 commands, and curated anti-patterns for sharp product thinking.

Why Acumen?

Impeccable taught AI to generate sophisticated frontend design. Acumen does the same for product management.

Every LLM learned from the same generic templates. Without guidance, you get the same predictable product slop: vague value props, demographic personas, vanity metrics, feature checklists with no thesis, and specs that could belong to any company.

Acumen fights that bias with:

  • A core skill with 7 domain-specific reference documents (view source)
  • 16 commands organized in 5 layers — from context gathering to shipping changelogs
  • 12 named anti-patterns ("AI Product Slop") that explicitly tell the AI what NOT to produce

What's Included

The Skill: product-thinking

A comprehensive product skill with 7 deep-dive references (view skill):

Reference Covers
Problem Discovery JTBD, Five Whys, Switch Framework, interview anti-patterns
User Segmentation Behavioral cohorts, personas, anti-personas, journey mapping
Prioritization & Strategy ICE/RICE limitations, sequencing logic, regret minimization
Metrics & Measurement Metric trees, counter-metrics, leading indicators, north stars
Specification Craft Working backward, acceptance criteria, edge case categories
Stakeholder Alignment DACI, disagree-and-commit, managing upward, pre-mortems
Competitive Intelligence Moat analysis, win/loss, competitive response playbooks

15 Commands in 6 Layers

Context Layer (build & maintain product knowledge)

Command What it does
/teach-acumen One-time setup that gathers product context and saves it to .acumen.md
/scout Build and maintain a living competitor map. Supports quick scans and deep analysis
/persona Build and maintain behavioral personas grounded in real user patterns
/features Build and maintain a lightweight feature inventory
/value Map per-persona value delivery to the usage and north star metrics that prove it

Audit Layer (assess product health)

Command What it does
/orientation Audit product identity — what we do, for whom, whether the feature surface is coherent with the thesis
/defensibility Audit what's hard to copy — moats, switching costs, data advantages, network effects, and AI-era resilience

Ideate Layer (find problems and opportunities)

Command What it does
/diagnose Find current problems based on data, value delivery for main personas, and current features. Suggests /workshop
/measure Check KPI health — what's working, what's not, where to dig deeper. Suggests /workshop
/workshop Ideation workshop to explore opportunities — value chain analysis, competitive gaps, user feedback synthesis

Craft Layer (build product artifacts)

Command What it does
/roadmap Plan a sequence of bets, break into shippable increments, prioritize
/increment Scope a product increment end-to-end — context, spec, implementation, wireframes, and engineering considerations. Leads to /critique-product
/critique-product Evaluate and validate a PM artifact — score for rigor, test riskiest assumptions, and decide if it's ready to ship

Communicate Layer (share with the world)

Command What it does
/narrate Write product communication for a specific audience, including when stakes are high and stakeholders disagree
/changelog Write a changelog from recent PRs and commits, grounded in feature context

Meta Layer (manage Acumen itself)

Command What it does
/update-acumen Fetch the latest Acumen skills from GitHub and update local copies
/cheatsheet Print a quick-reference cheatsheet of all Acumen skills, organized by layer

Usage Examples

/diagnose - Find the real problem

/diagnose why churn spiked in Q1
/diagnose onboarding drop-off

/orientation - Audit product identity

/orientation
/orientation after the pivot

/workshop - Ideate on opportunities

/workshop integration onboarding
/workshop expanding into enterprise

/increment - Scope a shippable increment

/increment self-serve team onboarding
/increment real-time collaboration on dashboards

/changelog - Write a changelog

/changelog last 7 days
/changelog since v2.1

/critique-product - Score and validate an artifact

/critique-product the PRD I just wrote
/critique-product our Q3 roadmap

Anti-Patterns: AI Product Slop

The skill includes 12 explicit tells of AI product slop:

  1. Template Filler - Every section filled, none grounded in what's actually built
  2. Demographic Persona - Demographics because real behavior is observed, not assumed
  3. Inverted Solution - Problem is the solution described negatively — the real friction was never observed
  4. Vanity Metric - Measures things that always go up, not what the product actually tracks
  5. Feature Checklist - Flat list with no awareness of what already exists or how it connects
  6. Generic Risk - "Technical complexity" — because the actual system was never examined
  7. Tautology Story - User story restates the feature because the real job-to-be-done is unknown
  8. Phantom Feature - Proposes what already exists — the codebase was never read
  9. Strategy of Everything - Lists every direction because without context, nothing can be ruled out
  10. Feature Roadmap - Timeline of features with no thesis about what the product is becoming
  11. Greenfield Spec - Scopes from scratch, ignoring existing architecture and shipped features
  12. Consensus Document - Inoffensive to all because without product evidence, there's nothing to argue from

Architecture

source/                          # Source of truth
  skills/
    product-thinking/            # Core skill (not user-invocable)
      SKILL.md
      reference/                 # 7 domain-specific reference docs
    teach-acumen/SKILL.md        # Context: setup
    scout/SKILL.md               # Context: competitors (+ deep analysis)
    profile/SKILL.md             # Context: personas
    features/SKILL.md            # Context: features
    value/SKILL.md               # Context: per-persona value & metrics
    update-acumen/SKILL.md       # Meta: sync skills from GitHub
    orientation/SKILL.md         # Audit: product identity
    defensibility/SKILL.md       # Audit: moats
    diagnose/SKILL.md            # Ideate: find problems
    measure/SKILL.md             # Ideate: KPI health check
    workshop/SKILL.md            # Ideate: opportunity exploration
    roadmap/SKILL.md             # Craft: sequence bets
    increment/SKILL.md           # Craft: scope increment
    critique-product/            # Craft: evaluate & validate
      SKILL.md
      reference/                 # Scoring rubric + reviewer personas
    narrate/SKILL.md             # Communicate: write for audience
    changelog/SKILL.md           # Communicate: write changelog

.agents/skills/                  # Agent Skills spec (generic)
.claude/skills/                  # Claude Code (symlinks to .agents/)

Context Files (per-project, gitignored)

File Maintained by Purpose
.acumen.md /teach-acumen Product context
.acumen/competitors.md /scout Living competitor map
.acumen/personas.md /persona Behavioral personas
.acumen/features.md /features Feature inventory
.acumen/value.md /value Per-persona value map with proof metrics
.acumen/value-chain.md /teach-acumen, /profile, /workshop Value chain map per persona
.acumen/sources.md /teach-acumen Data source configuration
.acumen/reports/ /diagnose, /measure, /workshop Point-in-time analysis reports

Installation

Claude Code

# Project-specific
cp -r .claude your-project/
cp -r .agents your-project/

# Or global
mkdir -p ~/.claude/skills
cp -r .agents/skills/* ~/.claude/skills/

Other AI Tools

Copy the .agents/skills/ directory into your tool's skill location. Skills follow the Agent Skills spec.

Usage

Once installed, start with setup:

/teach-acumen              # One-time: gather product context
/scout                     # Build competitor map
/persona                   # Build personas
/features                  # Inventory features

Then audit to understand where you are:

/orientation               # Is the product coherent?
/defensibility             # What's hard to copy?

Then ideate and build:

/diagnose                  # Find the real problem
/measure                   # Check KPI health
/workshop                  # Explore opportunities
/roadmap                   # Sequence the work
/increment                 # Scope an increment
/critique-product          # Score and validate before shipping

And communicate:

/narrate                   # Write for any audience
/changelog                 # Write a changelog

Design Principles

  1. Mindset injection > workflow templates - Skills set a thinking posture, not process steps
  2. Opinion > comprehensiveness - "RICE doesn't ship products, judgment does"
  3. Brevity > coverage - 50-100 lines per skill, depth in reference docs
  4. Context layer feeds everything - Not passive storage, active inputs
  5. Vocabulary calibration - Replace AI slop with product-fluent language
  6. Natural flow - Skills suggest the next skill: diagnose -> workshop -> increment -> critique

License

Apache 2.0. See LICENSE.

Inspired by Impeccable by Paul Bakaus.

About

Product Fluency for AI Harness

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors