Skip to main content
PromptMetricsPromptMetrics
Book a free call
← All open source
Skill

Frameworks fade. Habits stick.

You can sit through a great training, mean every word, and still slide back into old habits the first busy week. The Habit Layer gives the AI Fluency Framework's 4Ds a way to show up at the exact moment they matter: before Claude drafts, decides, analyzes, or produces something real, it asks you one direct question. Answer honestly, say skip, or just say "do it"; it proceeds either way.

MarkdownCC BY-NC-SA 4.0Claude Code · Claude.ai · Coworkgithub.com/promptmetrics/habit-layer
View on GitHubOpen an issue
Part of the Human Layer family, built by Ranya Barakat, same lineage as the coaching skill. Applies the AI Fluency Framework (Rick Dakan, Joseph Feller, and Anthropic). Reflection prompts only, not legal, compliance, or policy advice.
Install

Claude Code

Model-invoked once installed; also runnable directly.

claude
> /plugin marketplace add promptmetrics/habit-layer
> /plugin install the-habit-layer@habit-layer
# update later: /plugin marketplace update habit-layer

Claude.ai / Cowork

Requires Code execution and file creation enabled under Settings ▸ Capabilities.

bash
# Customize ▸ Skills → + Create skill → Upload a skill
# upload the-habit-layer.skill from the repo, then toggle it on
What you get
  • One question, well timed: fires when a real task looks vague, high-stakes, or irreversible. Never twice on the same task.
  • Mapped to the 4Ds: each question targets whichever of Delegation, Description, Discernment, Diligence is most exposed.
  • Never a gate: "do it" always proceeds; it stays quiet if your message already answers the question.
  • Atomic Habits mechanics: small cues, low friction, real reward for engaging: behavior change, not compliance.
  • Stands alone: no course required; pairs with the PromptMetrics AI Fluency Cohort for the deeper version.
How it works

A cue at the moment of delegation, not a lecture after the fact.

  1. 01

    Watch

    The skill monitors for real tasks (drafting, deciding, analyzing) headed somewhere that matters.

  2. 02

    Map

    It picks the most exposed of the 4Ds for this moment: Delegation, Description, Discernment, or Diligence.

  3. 03

    Ask once

    One direct question. "Is this heading to a client? Would you sign it exactly as is?"

  4. 04

    Proceed

    Answer, skip, or "do it"; the task runs either way. The habit compounds; the friction doesn't.

Try it
"Draft the renewal email to Northwind."→ "Is this heading to a client? Would you sign it exactly as is?"
"Analyze this churn data."→ "What does a strong outcome actually look like here, in your own words?"
/the-habit-layer:the-habit-layerRun it directly in Claude Code.

Questions or issues? Open one on GitHub, or browse the rest of the catalog.

All open source