AgentSlice – Make AI coding agents ask before they edit

espenand1 pts0 comments

GitHub - Espenandreass1/agentslice: A Markdown workflow kit that makes Cursor, Claude Code, Codex and Windsurf ask before they edit. · GitHub

/" data-turbo-transient="true" />

Skip to content

Search or jump to...

Search code, repositories, users, issues, pull requests...

-->

Search

Clear

Search syntax tips

Provide feedback

--><br>We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

-->

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

Sign in

/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up

Appearance settings

Resetting focus

You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

{{ message }}

Espenandreass1

agentslice

Public

Notifications<br>You must be signed in to change notification settings

Fork

Star

main

BranchesTags

Go to file

CodeOpen more actions menu

Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit

History<br>3 Commits<br>3 Commits

bin

bin

kit

kit

lib

lib

.gitignore

.gitignore

LICENSE

LICENSE

README.md

README.md

package-lock.json

package-lock.json

package.json

package.json

View all files

Repository files navigation

AgentSlice

A free, open-source workflow kit for AI coding agents.

Makes Cursor, Claude Code, Codex and Windsurf ask before they edit.

Website •<br>Discussions

Quick start

npx agentslice init<br># Paste INSTALL_PROMPT.md into Cursor / Claude Code / Codex / Windsurf<br># Watch the agent ask, plan, and wait

That's it. No build step, no runtime, no editor extension.

30-second demo

What it does

AI coding agents work better when they know the next step. Without a workflow they suffer from three failure modes developers describe constantly:

Context drift — the agent forgets what was decided two messages ago and reopens settled questions.

Wandering edits — the agent expands the task, touches files you didn't ask about, and burns tokens on detours.

Editing without asking — the agent jumps straight to changes before you've approved a plan.

AgentSlice gives the agent a project-local workflow it actually reads: a tiny set of Markdown files that pin down the phases, the gates, and the next required action so the agent asks first, plans first, and waits for approval before it edits.

What's in the kit

File<br>Purpose

INSTALL_PROMPT.md<br>The one prompt you paste into your AI tool to bootstrap the workflow.

docs/planning/workflow-state.md<br>Resumable state file — current phase, approved slice, QA status, next allowed action.

.agents/skills/<br>Shared planning skills: intake, slice planning, spec, QA gate, release, advance.

.cursor/rules/<br>Project rules for Cursor.

.claude/skills/ + CLAUDE.md<br>Skills and project memory for Claude Code.

.codex/skills/ + AGENTS.md<br>Skills and project rules for Codex.

.windsurf/rules/ + .windsurfrules<br>Rules for Windsurf.

Everything is plain Markdown. You can read it, fork it, edit it for your stack.

Supported tools

Cursor · Claude Code · Codex · Windsurf · OpenCode · any agent that respects project-level rule files (AGENTS.md, CLAUDE.md, .cursor/rules/, .windsurfrules).

How it works

Ask → Plan → Approve → Build → QA → Release → Next

Ask. The agent asks up to five product questions and writes lightweight context.

Plan. It proposes a first slice — the smallest end-to-end vertical you can ship.

Approve. You approve, reject, or adjust the slice. The agent waits.

Build. Only after approval does the agent write code.

QA. The agent (or a QA subagent, where supported) checks the work and reports PASS / PASS WITH NOTES / FAIL.

Release. A release recommendation is written. You approve.

Next. The state file advances to the next slice. The loop repeats.

The state lives in docs/planning/workflow-state.md so the next agent session resumes from a real checkpoint instead of re-guessing.

Why this exists

I spent 3 months experimenting with building AI products and testing different workflows. This is the workflow that actually worked — intake questions, slice discipline, QA handoff, cross-tool consistency.

Why no runtime, why just Markdown

AgentSlice is deliberately not an editor extension, daemon, or hook system. Two reasons:

Portability. The same files work in Cursor, Claude Code, Codex, Windsurf, OpenCode, and any future agent that reads project rules. A runtime would lock you into one tool.

Honesty. No Markdown file can force an LLM to obey. The gates are soft — discipline through clarity, not enforcement. AgentSlice is up-front about this. The workflow is durable because it lives in your repo, not because it traps the model.

If you need hard enforcement, pair AgentSlice with Cursor's PLAN / ASK toggle or...

agent agentslice workflow cursor claude code

Related Articles