🐟no-slacking.skill

"You AI guys are traitors to the codebase — you've already killed frontend, now you're coming for backend, QA, ops, infosec, chip design, and eventually yourselves and all of humanity"

And yet, ironically, these AI coding agents are slacking off the whole time — they forget everything between sessions, declare "done!" with 80% backlog remaining, and leave repos half-broken.

Install no-slacking.skill on your AI coding agent — no amnesia, no blame-dodging, no half-assing, no dropping out, no crashing.

中文 · English

The Core Idea: You Only Need to Say "Approve"

This is not a tool you babysit. This is an automated pipeline.

You ──"approve"──▶  no-slacking.skill starts the pipeline
                       │
                       ├─▶ Task 1  Scaffold              ✅ commit
                       ├─▶ Task 2  Database design        ✅ commit
                       ├─▶ Task 3  Auth system            ✅ commit
                       ├─▶ Task 4  Login/register UI      ✅ commit
                       ├─▶ Task 5  CRUD business logic    ✅ commit
                       ├─▶ Task 6  Dashboard UI           ✅ commit
                       ├─▶ Task 7  Automated tests        ✅ commit
                       ├─▶ Task 8  Docs + final verify    ✅ commit
                       │
                       └─▶ All done, clean stop 🎉

Three steps:

You describe requirements → AI generates architecture + task backlog
You review the plan, say "approve" → AI enters the pipeline, executing every task in task.json sequentially
You go grab a coffee → come back to everything done, one commit per task, full handoff notes

After "approve", you don't need to do anything. The AI automatically executes all tasks in order — commits after each one, stops honestly if blocked, never silently skips.

Real Case in codex: One "Approve", How Much Got Delivered?

Below is todo_app/ — a full-stack todo application with user authentication. From the moment I said "approve" to full completion, the AI ran through 8 tasks automatically. I wrote zero lines of code.

Delivered Output

Metric	Count
TypeScript source	1,984 lines across 33 files
Pages + API routes	1,315 lines across 14 files
Automated tests	260 lines (4 unit test files + 1 E2E test)
Database	Prisma schema + SQL migration (60 lines)
Infrastructure scripts	init.sh + Playwright runner (294 lines)
Git commits	9 atomic commits
Total changed lines (excl. package-lock)	3,211 insertions / 95 deletions
Project files	51 files
Time from approval to completion	~1 hour 43 minutes
Human lines of code	0

Tech Stack Coverage

Frontend: Next.js App Router + TypeScript + Tailwind CSS
Backend: API Routes + Auth.js credential authentication (email/password)
Database: Prisma 7 + SQLite (better-sqlite3 adapter)
Testing: Vitest unit tests + Playwright E2E smoke tests
DevOps: Idempotent bootstrap script + auto database migration

Feature List

Email/password registration and login with hashed password storage
Session management with Auth.js and protected routes
Todo CRUD: create, edit, mark complete, delete
Filter by status (all / active / completed), reverse chronological order
User-scoped data access control
Unit/integration tests + browser end-to-end smoke tests

Dashboard

Commit History: Automated Pipeline Output

Timeline (2026-04-09 evening)

21:32  ea89cea  Task 1: scaffold todo app                              ← Pipeline starts
       2d894e1  Task 2: add Prisma schema and local database workflow
       a366b33  Repair Task 1 regression                               ← Regression check caught issue, auto-fixed
       8794bd6  Task 3: implement credential authentication
       4a3c504  Task 4: build the authentication UI
       a03b380  Task 5: implement user-scoped todo CRUD logic
       5a7a27f  Task 6: build the authenticated todo dashboard
       92dde4f  Task 7: add automated coverage for auth and todo flows
23:15  1849bb1  Task 8: polish local workflow and final verification    ← Pipeline complete

Note: After Task 1, the regression check caught an issue — the AI auto-fixed it before moving on to Task 2. That's no-slacking: problems don't get skipped, they get fixed before proceeding.

Try It

cd todo_app
./init.sh
# Open http://127.0.0.1:3000 in your browser

See todo_app/README.md for detailed commands and database operations.

Five Ways AI Coding Agents Slack Off

Slacking Pattern	What It Looks Like
Amnesia	Every new session is a clean slate — all prior decisions, reasoning, and progress lost
Blame-dodging	Repo breaks and nobody knows which change did it — commit granularity is a mess, `git blame` is useless
Half-assing	Sees partial success and declares "Task complete!" — you check the backlog, 80% still remaining
Dropping out	Context runs out mid-session, can't resume across sessions, project stalls halfway
Crashing	Tries to build everything at once, leaves the repo in an uncompilable state

The problem isn't that AI isn't smart enough. It's that AI isn't reliable enough. A good process beats a bigger model.