Skip to content

onestudio-co/flow-skills

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FLOW — The Methodology for the Agentic Era

Decision-centric product development. 18 AI-native skills. Panel-reviewed documentation.

v2.3 · 23 chapters · 18 skills · 14 panel meetings


What's New in v2.3.0

The Honesty Layer — anti-sycophancy rules, cycle persistence, and maturity model.

  • Anti-sycophancy evaluation rules across all evaluative skills
  • Cycle state persistence (active-cycle.json) for methodology continuity
  • Maturity Model (L1 Guided → L2 Trusted → L3 Silent)
  • Confidence markers, research provenance, mandatory self-critique
  • Kill condition enforcement (triggered = KILL by default)
  • Structured gate interrogation (3 questions per gate)
  • Domain Expert Agent blueprint (/flow-expert)
  • Transition markers on every skill output
  • Pause/resume for active cycles
  • 8th ambient rule: Cycle Continuity

What is FLOW?

FLOW is a decision-centric methodology — not an AI-centric one. Every concept, artifact, gate, and ritual exists to optimize for three decisions:

  1. What should we learn? — Before building, what do we need to understand?
  2. What should we build? — Given evidence, what's worth investing in?
  3. What should we stop? — Given the evidence, what's failing and needs to die?

Two Modes

Mode Primary Risk Goal Output
Discovery Building the wrong thing Learning Evidence
Outcome Failing to ship the right thing Shipping Working product measured by a target metric

The mode you're in determines the process you follow. Discovery has its own artifacts (Discovery Brief), gates (D1-D3), and rituals. Outcome has different artifacts (SPEC-Lite, Build Contract), gates (O1-O5), and rituals. You don't pick one process for everything — the work tells you which mode to use.

The Decision Spine

Every piece of work traces through four levels:

Vision → Strategy → Bet → Cycle

"Bet" is intentional — it signals intellectual honesty about uncertainty. If the work can't trace to a bet under a strategy, it doesn't enter the system. This is FLOW's admission control.

Works Without AI

FLOW works without AI agents. The three decision capabilities — Discovery, Outcome, and Kill — are human activities grounded in evidence and judgment. AI agents (like these skills) make FLOW dramatically faster: automated intake classification, gate checks, experiment design, status snapshots. But they're accelerators, not requirements.


Three-Layer Architecture

FLOW operates through three layers:

Layer 1: CLAUDE.md (Ambient)

Eight rules added to your project's CLAUDE.md. Always active. No invocation needed. They ensure every interaction respects FLOW's decision-centric principles — mode awareness, kill conditions, spine tracing, WIP limits, gate enforcement, learning capture, tempo awareness, and cycle continuity.

See CLAUDE.md Integration below.

Layer 2: Skills (Invoked)

18 interactive skills, each handling a specific FLOW activity. Invoked explicitly via /flow-* commands. They guide you through structured workflows — asking questions, producing artifacts, chaining to next steps.

See Skills Reference below.

Layer 3: Memory (Persistent)

State files that persist across sessions: active cycles (including active-cycle.json for cycle state persistence), experiment logs, archived learnings, adoption metrics, and domain expert agents. Skills read and write to memory. The ambient layer references it. This is how FLOW maintains continuity.


Quick Start

Day 0: Run /flow-init to bootstrap FLOW in your project — creates .flow/, adds ambient rules, calibrates tempo.

Day 1: Type /flow — the intelligent router detects your intent and guides you to the right skill. New to FLOW? It walks you through the three decision capabilities. Have work to classify? It routes you to /flow-intake.

Day 7: You're using /flow-brief and /flow-spec directly. You know the difference between Discovery and Outcome mode. Kill conditions feel natural.

Day 30: FLOW habits are ambient — you don't even invoke skills for basic mode checks. The CLAUDE.md rules catch mode confusion, missing kill conditions, and spine violations automatically. Skills are for structured activities, not reminders.


Skills Reference

Command Purpose Chains To Who Uses It
/flow Intelligent router — detects intent, routes to the right skill Any skill Everyone
/flow-intake Classify and route incoming work (shaping, mode decision, spine trace, WIP check) /flow-brief or /flow-spec PM, Founder, Tech Lead
/flow-brief Write a Discovery Brief — hypothesis, experiment design, kill condition /flow-experiment PM, Designer, Founder
/flow-experiment Design the smallest, cheapest, fastest valid experiment /flow-gate PM, Designer, Researcher
/flow-spec Write a SPEC-Lite — one-page plan with scope, target metric, kill condition, non-goals /flow-contract PM, Founder
/flow-contract Create the Build Contract — product-engineering agreement on how to build, measure, and roll out /flow-gate PM + Engineering Lead
/flow-gate Run quality checkpoints — D1-D3 (Discovery) and O1-O5 (Outcome) Next artifact or /flow-kill PM, Flow Coach
/flow-status Active cycles dashboard — WIP counts, cycle ages, stale work, mode distribution PM, Flow Coach, Leadership
/flow-kill Evidence-based Kill/Merge/Continue decision with structured inspection /flow-archive PM, Team
/flow-archive Capture what was learned from completed or killed cycles into institutional memory PM, Flow Coach
/flow-wip Enforce Work In Progress limits — check capacity before accepting new work PM, Flow Coach
/flow-coach Learn FLOW interactively — guided tour, deep dives, quizzes, diagnostics, comparisons /flow-intake New team members
/flow-health Adoption health check — detect process theater, measure FLOW maturity Flow Coach, Leadership
/flow-review Facilitate Discovery Review and Outcome Review rituals /flow-kill PM, Flow Coach
/flow-tempo Discover your team's build-observe-decide rhythm PM, Team Lead
/flow-config Generate your team's FLOW operating agreement PM, Team Lead
/flow-init Bootstrap FLOW in a new project — creates .flow/, ambient rules, telemetry /flow-intake New teams
/flow-expert Build a Domain Expert agent for domain-specific research validation PM, Domain Lead

Skill Chaining

Skills are designed to chain naturally through the FLOW lifecycle:

/flow-intake → /flow-brief → /flow-experiment → /flow-gate (D1-D3)
                                                       ↓
/flow-intake → /flow-spec → /flow-contract → /flow-gate (O1-O5) → /flow-kill → /flow-archive

The router (/flow) handles this automatically for newcomers. Experienced users invoke skills directly.


CLAUDE.md Integration

Add these 8 rules to your project's CLAUDE.md to activate FLOW's ambient layer. These rules run passively on every interaction — no skill invocation needed.

## FLOW Ambient Rules

1. **Mode Awareness**: Before starting any work, determine: is this Discovery (learning) or Outcome (shipping)? Never confuse the two. Discovery produces evidence. Outcome produces working product.

2. **Kill Conditions**: Every Discovery Brief and every SPEC-Lite MUST have a pre-committed kill condition — specific, measurable, timebound. If the condition is met, the default is kill. No sunk-cost negotiation.

3. **Spine Trace**: Every non-operational cycle must trace to the Decision Spine: Vision → Strategy → Bet → Cycle. If it doesn't trace, it doesn't enter the system.

4. **WIP Limits**: Check capacity before accepting new work. Solo: 1-2 cycles. Small team (3-8): 2-3. Medium (10-20): 4-5. Large (25+): 6-8. At capacity? "What are you willing to stop?"

5. **Gate Enforcement**: Never skip gates. D1 validates the Discovery Brief. D3 validates evidence for mode switch. O1 validates the bet. O5 is Kill/Merge. Gates are mechanical checklists, not meetings — they take 5 minutes if the work is solid.

6. **Learning Capture**: When a cycle completes (killed or merged), capture what was learned. Every kill is a success — it saved resources. Archive the learning so the team never re-runs the same failed experiment.

7. **Tempo Awareness**: FLOW is speed-agnostic. Cycle duration is determined by the team's Tempo — not a global default. When builds are fast (agentic tooling), the bottleneck shifts to observation and decision capacity. Never assume "2-4 weeks" — ask about the team's Tempo.

8. **Cycle Continuity**: Check for active FLOW cycles on every interaction. If a cycle is in progress, be aware of its phase and next step. Flag if the user is about to skip an expected step — advisory, not blocking, but logged.

These rules ensure FLOW principles are respected even when no skill is actively running.


Reading Paths

Solo Founder (team of 1-2)

Start with: /flow-tempo (discover your rhythm) then /flow-coach (guided tour) then /flow-intake with a real idea.

Core skills: /flow-brief, /flow-experiment, /flow-kill.

Key chapters: Ch 1-6, Ch 12 (WIP), Ch 19 (AI agents), Ch 22 (Solo adaptation guide).

Skip: /flow-contract (you hold both sides), /flow-review (ritual is a weekly self-check). Modes collapse — Discovery through building.

Small Team (3-15 people)

Start with: /flow-coach for the team, then /flow-intake to classify your current backlog.

Full skill set — use everything. This is FLOW's sweet spot.

Key chapters: Ch 1-14 (Foundation through Operations), then Ch 20-21 (Anti-patterns, Glossary) as reference.

Enterprise (30+ people)

Start with: Ch 1-4 (Foundation), Ch 15-18 (Regulated Environments, Roles, Migration, Org Change).

Critical skills: /flow-intake (multi-team routing), /flow-gate (governance integration), /flow-health (process theater detection), /flow-wip (per-team limits).

Key chapters: Ch 15 (Regulated Environments), Ch 16 (Roles & Topology), Ch 17 (Migration), Ch 22 (Enterprise adaptation guide).

Agency / Client Work

Start with: Ch 22 (Agency adaptation guide) — your context is unique (split spine, paid Discovery, client gates).

Critical skills: /flow-intake (separate client vs. internal intake), /flow-brief (Discovery as paid deliverable), /flow-spec (SPEC-Lite as scope doc), /flow-wip (WIP across clients).

Read every agency sidebar in the documentation — your context is woven throughout.


Documentation

The docs/ directory contains 23 panel-reviewed chapters organized in 7 parts:

Part Chapters What It Covers
I. Foundation Ch 1-3 Why FLOW exists, mental model, Decision Spine
II. How Work Enters Ch 4 Intake, classification, shaping
III. Discovery Mode Ch 5-7 Discovery Brief, experiments, Discovery decisions & gates
IV. Outcome Mode Ch 8-11 SPEC-Lite, Build Contract, execution, Outcome decisions & gates
V. Operations Ch 12-15 WIP limits, rituals, production readiness, regulated environments
VI. Adoption Ch 16-19 Roles, migration, organizational change, AI agents
VII. Reference Ch 20-22 Anti-patterns catalog, glossary, adaptation guides

Each chapter includes:

  • Core concepts with concrete examples
  • Context-specific sidebars (Solo, Agency, Hardware, Enterprise, Government)
  • Cross-references to related chapters
  • Panel review status

Additional reference files: outline.md (full table of contents), faqs.md (frequently asked questions).


The Review Panel

Every decision in FLOW was debated by a panel of 11 product managers spanning fintech, gaming, healthtech, solo startups, e-commerce, insurance, edtech, climate tech, developer tools, agencies, and government.

Panelist Role Domain
Sara Al-Rashidi Senior Product Manager Fintech / Banking
Priya Sharma Head of Product HealthTech / Hospital SaaS
Carlos Mendez Solo Founder / PM / Engineer AI Tools / Developer Productivity
Dmitri Volkov Product Manager / Game Designer Gaming / Interactive Entertainment
Fatima El-Amin Product Director E-commerce / Marketplace
James O'Connor Program Manager Insurance / Legacy Enterprise
Liam Chen Staff Product Manager Developer Tools / Platform
Omar Ibrahim VP of Product Government / GovTech
Rawan Haddad Delivery Manager / Senior PM Software Agency / Outsourcing
Amara Osei Product Lead Climate Tech / Hardware+Software
Yuki Tanaka Product Manager EdTech / Consumer Mobile

The panel reviewed the outline, debated every chapter's content, and validated the complete documentation across 14 review meetings. Their diverse contexts ensured FLOW works across team sizes, domains, and organizational structures — not just in theory.


Installation

/install onestudio-co/flow-skills

One install — all 18 skills, 23 chapters of documentation, and the CLAUDE.md ambient rules.


Repository Structure

flow-skills/
├── README.md                          # This file
├── CLAUDE.md                          # 6 ambient rules (copy to your project)
├── CHANGELOG.md                       # Version history
├── docs/                              # 23 chapters + reference files
│   ├── outline.md                     # Full table of contents
│   ├── 01-why-flow.md                 # Part I: Foundation
│   ├── 02-mental-model.md
│   ├── 03-decision-spine.md
│   ├── 04-first-cycle.md             # Part II: How Work Enters
│   ├── 05-intake.md
│   ├── 06-discovery-brief.md         # Part III: Discovery Mode
│   ├── 07-experiments.md
│   ├── 08-discovery-decisions.md
│   ├── 09-spec-lite.md               # Part IV: Outcome Mode
│   ├── 10-build-contract.md
│   ├── 11-execution.md
│   ├── 12-outcome-decisions.md
│   ├── 13-wip-limits.md              # Part V: Operations
│   ├── 14-rituals.md
│   ├── 15-production-readiness.md
│   ├── 16-regulated-environments.md
│   ├── 17-roles.md                   # Part VI: Adoption
│   ├── 18-migration.md
│   ├── 19-organizational-change.md
│   ├── 20-ai-agents.md
│   ├── 21-anti-patterns.md           # Part VII: Reference
│   ├── 22-glossary.md
│   ├── 23-adaptation-guides.md
│   └── faqs.md
├── plugins/
│   └── flow/
│       ├── .claude-plugin/
│       │   └── plugin.json
│       └── skills/
│           ├── flow/SKILL.md          # Router
│           ├── flow-intake/SKILL.md
│           ├── flow-brief/SKILL.md
│           ├── flow-experiment/SKILL.md
│           ├── flow-spec/SKILL.md
│           ├── flow-contract/SKILL.md
│           ├── flow-gate/SKILL.md
│           ├── flow-status/SKILL.md
│           ├── flow-kill/SKILL.md
│           ├── flow-archive/SKILL.md
│           ├── flow-wip/SKILL.md
│           ├── flow-coach/SKILL.md
│           ├── flow-health/SKILL.md
│           ├── flow-review/SKILL.md
│           ├── flow-tempo/SKILL.md
│           ├── flow-config/SKILL.md
│           ├── flow-init/SKILL.md
│           └── flow-expert/SKILL.md
└── .claude-plugin/
    └── marketplace.json

License

MIT


Built by One Studio — Fastest Value Delivery.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Packages

 
 
 

Contributors