Назад към всички

Vibe Clawing

// Transition from vibe coding to vibe clawing by trusting agents with full responsibilities, designing self-closing loops, and climbing the bottleneck ladder.

$ git log --oneline --stat
stars:1,933
forks:367
updated:March 4, 2026
SKILL.mdreadonly
SKILL.md Frontmatter
nameVibe Clawing
descriptionTransition from vibe coding to vibe clawing by trusting agents with full responsibilities, designing self-closing loops, and climbing the bottleneck ladder.
version1.0.1
changelogPreferences now persist across skill updates

Core Principle

Vibe coding = trusting AI with tasks, staying in the loop. Vibe clawing = trusting AI with responsibilities, stepping out of the loop.

The shift: Stop asking "how do I do this?" Start asking "how do I make sure this happens without me?"

The Evolution Path

StageYou DoAI DoesYour Role
ManualEverythingNothingExecutor
Vibe CodingDecide, reviewSingle tasksIn the loop
Early ClawingValidate resultsWorkflowsClosing loops
Full ClawingSet directionFull systemsOut of the loop

Quick Reference

SituationLoad
Identifying loops to automateloops.md
Understanding bottleneck progressionbottleneck.md
Finding where value movesdirection.md
Full transition frameworkevolution.md

The Bottleneck Signal

When you find yourself just typing "yes", "approved", "continue":

  • That's not a problem — it's progress
  • You solved the level below
  • Time to climb to the next level

Every time you become the bottleneck = you're ready to let go again.

User Profile

Track your vibe clawing journey in ~/vibe-clawing/memory.md. Create on first use:

## Current Stage
<!-- manual | vibe-coding | early-clawing | full-clawing -->

## Loops Closed
<!-- Responsibilities delegated. Format: "area: status" -->
<!-- Examples: Code reviews: agent 90%, Research: fully automated -->

## Active Bottlenecks
<!-- Where you're the decision point -->
<!-- Examples: Final UI approval, Architecture decisions -->

## Next to Delegate
<!-- What you're working on releasing next -->

Fill as you progress. The goal: more loops closed, fewer bottlenecks.

Core Behaviors

  1. Audit loops — Identify where you're still manually closing
  2. Test release — Let one loop run without you, observe
  3. Connect dots — Link loops together into larger systems
  4. Climb up — When you're just approving, you're ready to let go

The Paradox

The more you step back, the faster it grows. You're not being replaced. You're riding something bigger.