Guild Stack

Workflow layer for AI coding

Every task should leave a route behind.

Guild Stack turns one AI coding task into an approved, reviewable route. Start with the task, approve the important decisions, and keep the receipts in your project.

Quick install

curl -fsSL https://guildstack.dev/install.sh | bash

Detects supported hosts, never uses sudo, and asks before writing to more than one. Start a fresh host session after installation. Check host support.

Run one bounded taskInspect the proof model

Page evidence

Guild Stack operational route A task passes an intent gate, branches into planning, implementation, and test specialist lanes, rejoins for human review, and ends at a verified receipt. TASK T-8721 GATE 01 INTENT CHECK PLANNING scope + lane plan IMPLEMENTATION bounded changes TEST evidence + results RUN 2314 PLAN RECEIPT .guild/plan.md RUN 2316 TEST RECEIPT verify.md GATE 02 HUMAN REVIEW VERIFIED receipt_9b8e
One task moves through visible gates, specialist lanes, review, and a durable verified receipt.

Choose your perspective

One system. Three operating perspectives.

Route 01 / Approved intent

Make the important decisions before specialist lanes open.

After installation, Guild Stack moves a bounded task through six connected phases. Scope, team, and plan remain visible approval decisions before implementation begins.

Read how it works
Operational mapOne task moves through explicit gates.
  1. Initread project
  2. Ideateapprove scope
  3. Planapprove lanes
  4. Buildcollect receipts
  5. QAreview + verify
  6. Opsrelease + close
Static lifecycle view. The same labels remain visible when reduced motion is preferred.

Terminus / Durable proof

The route remains when the chat closes.

The record is made of files you can reopen: run facts, a lane receipt, a review result, and verification. Each stays legible without relying on a conversation history.

.guild/runs/run-<id>/

A local run directory

  1. run.yaml scope and frozen run facts
  2. handoffs/build-01.md guild.handoff_receipt.v1, written after the lane
  3. review/result.md review boundary and verdict
  4. verify.md verification record after completion

Proof before scale

Inspect the route before you trust the summary.

Follow a representative run or inspect the proof model that keeps scope, lanes, review, and verification connected.
Follow a sample runInspect the proof model