AI is powerfulβbut not magic. Teams shipping with AI today face a hard truth: output quality depends entirely on clarity. You need to know what to delegate, how to describe it, how to verify it works, and who's accountable. That's the 4D Framework: Delegation, Description, Discernment, Diligence.
Fluently operationalizes the 4D Framework. It gives you a CLI, MCP server, and a growing knowledge base of best practicesβbuilt on Anthropic's AI Fluency research by Dakan & Feller. Score your team's AI workflows. Learn from the community's Fluently 4D cycles. Ship with confidence.
Real-world example: Compare your code review process against community best practices:
$ fluent compare --task "Automated code review with human sign-off"
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β TOP 3 SIMILAR FLUENTLY 4D CYCLES FROM KNOWLEDGE BASE β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β 1. Code Review: Review Depth vs. Speed Tradeoffs β
β Delegation: 75 Description: 82 Discernment: 78 β
β Diligence: 71 OVERALL: 76/100 β
β β
β 2. Bug Fix Prioritization β
β Delegation: 68 Description: 74 Discernment: 71 β
β Diligence: 69 OVERALL: 71/100 β
β β
β 3. Test Case Generation β
β Delegation: 72 Description: 76 Discernment: 70 β
β Diligence: 68 OVERALL: 72/100 β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββEach suggestion links to the full Fluently 4D cycleβdelegation guidelines, framing prompts, hallucination patterns to watch for, and approval workflows.
Option 1 β Zero install (browser): Use the live scorer right now. No setup required.
Option 2 β Zero install (terminal): Run instantly with npx β no PATH changes needed:
npx fluently-cli score "Automated bug triage with human review"Option 3 β Permanent install:
npm install -g fluently-cli
fluent --helpRequires Node.js 20+. The CLI is self-contained β no configuration files needed.
CLI β Run fluent in your terminal to score 4D workflows, compare against community patterns, and contribute new Fluently 4D cycles.
MCP Server β Embed 4D scoring into any AI agent or MCP-compatible client (VS Code Copilot, Cursor, Continue, Claude Desktop, Claude Code, and more). Expose tools like find_relevant_cycles(), get_dimension_guidance(), and contribute_cycle() to your agent.
Knowledge Base β 50+ community-contributed Fluently 4D cycles organized by domain (coding, writing, research, management, product, etc.). Each cycle is scored on the 4D dimensions and peer-reviewed before merge.
This isn't just a frameworkβit's a commons for AI fluency. Every Fluently 4D cycle you contribute teaches thousands of teams how to collaborate smarter with AI.
Built on the AI Fluency 4D Framework by Dakan & Feller. Everything β code, knowledge cycles, and tooling β is released under the MIT License.
- Install the CLI and run
fluent compare --task "your workflow"to see how others solve similar problems. - Check the Fluently 4D cycles β find your use case in the knowledge base above.
- Score your process β run
fluent scorewith your delegation intent to identify gaps.
- Integrate the MCP server into your AI agent or IDE plugin (any MCP-compatible client).
- Expose 4D scoring to your users β let them verify AI output quality before shipping.
- Use the scorer engine directly β
const { scoreTask } = require('@fluently/scorer').
Share your Fluently 4D cycles and help the community. We're looking for patterns, anti-patterns, and real-world lessons from teams shipping AI features.
β Contributing Guide β Detailed walkthrough for submitting a new Fluently 4D cycle.
Every task, workflow, or prompt gets scored across four dimensions:
| Dimension | Question | What We Measure |
|---|---|---|
| Delegation | Who decides? | Clarity on AI autonomy vs. human oversight. Escalation triggers matter. |
| Description | What context? | Quality of framing + examples + constraints. Reduces ambiguity. |
| Discernment | Is it right? | Your ability to spot hallucinations and overconfidence. Red flags & green signals. |
| Diligence | Who's accountable? | Governance, review workflows, audit trails. Who approves before shipping. |
Each scores 0β100. The framework doesn't replace human judgmentβit sharpens it.
Embeddings-based similarity β Replace keyword matching with semantic search so find_similar_cycles() surfaces truly relevant patterns regardless of terminology.
VS Code extension β Inline 4D scoring in your editor. @fluently in comments to get suggestions while you write, review, or debug.
Web playground β Try 4D scoring live without installing. Experiment with Fluently 4D cycles. Generate your own.
CLI β Commander.js + multi-provider AI SDK
MCP Server β Model Context Protocol (stdio)
Scorer β Zod schema validation + keyword matching
Knowledge β YAML + JSON + GitHub API
Tests β Vitest
Language: TypeScript β’ Runtime: Node.js 20+ β’ Module: ESM
- Share a Fluently 4D cycle β Takes 15 minutes. Write YAML. Open a PR. CI validates schema.
- Improve the scorer β Suggest semantic improvements, new dimensions, new domains.
- Build an integration β MCP server is stable. Write a Slack app. A GitHub Action. A web service.
By contributing, you help us scale the standard. Every Fluently 4D cycle in the knowledge base teaches thousands of teams.
- π Live Site & Knowledge Browser
- π Full Documentation
- π¬ GitHub Discussions
- π Report Issues
- π€ Contributing Guide
The AI Fluency 4D Framework was developed by Dakan & Feller as a collaborative model for operationalizing AI fluency.
Fluently brings that framework to life as an open-source tool + knowledge commons.
- Framework: Dakan & Feller β AI Fluency 4D Framework
- License: MIT (code + knowledge)
MIT β code, knowledge cycles, and all tooling.
Mix and match. Share freely. Build better AI collaboration.
Happy shipping. π