AI coding workflow

A workflow you can steer, challenge, and resume.

Start with one bounded task. Guild Stack connects approved intent, focused execution, review, verification, operations, and project memory in one route you can reopen.

Page evidence

Capability system

map / active

Every capability ends at inspectable state.

  1. 01
    Decide

    Scope, team, and lane plan

    approved intent
  2. 02
    Execute

    Focused context and specialist work

    handoff receipt
  3. 03
    Challenge

    Review the current artifact and verify

    review result + verify
  4. 04
    Remember

    Promote durable knowledge and recall it

    wiki + context refs
  5. 05
    Adapt

    Select host capability and record loss

    degradation receipt
The product system joins decisions, execution, challenge, memory, and host adaptation through durable evidence.

Lifecycle spine

Every feature follows the same accountable path.

  1. 01Init

    Read the project and memory before proposing scope.

  2. 02Ideate

    Turn the ask into approved goals and criteria.

  3. 03Plan

    Compose the phase team and lane plan.

  4. 04Build

    Run bounded specialist lanes with handoff receipts.

  5. 05QA

    Challenge, verify, and compute release readiness.

  6. 06Ops

    Record runbooks, rollback, and closure evidence.

02 / Decide and execute

Give specialists the context they need, then collect the receipt.

Scope, team, and plan remain visible before work begins. Each specialist receives a bounded task and finishes with evidence the next lane can reopen.

Operational mapAssignments go out. Typed receipts come back.
Lead
  1. lane assignment
  2. review packet
  3. gate decision
Specialist
  1. focused context
  2. bounded work
  3. handoff receipt
.guild/runs/
  1. task
  2. handoff
  3. review result

Files carry authority, evidence, and recovery state. Chat is not the durable bus.

Execution stays bounded because assignments and receipts use the project-local file bus.

03 / Challenge and prove

A result is not done until the current artifact passes.

Review, verification, release readiness, and operations remain separate gates. The result is checked against the current artifact, and review strength stays visible.

Review decision

map / active

Challenge the bound artifact, then route the verdict.

  1. 01 / PacketBind scope + artifactChecksum, evidence, reviewer posturereview packet
  2. 02 / ChallengeTest the current workCross-host when available; weaker modes recordedreview result
  3. 03 / DecideBlock or satisfyFindings route to rework; satisfaction advancesverdict + trail
  4. 04 / VerifyProve doneRun checks against the final diffverify.md
Blocking findings return to bounded rework. A satisfied result advances to verification with the review trail intact.

04 / Remember, control, adapt

Go deeper only where the job demands it.

Operational mapMemory, control, and adaptation remain connected to the run.
Remember
  1. run evidence
  2. human promotion
  3. canonical wiki
Control
  1. operator setting
  2. frozen posture
  3. resolved settings
Adapt
  1. host capability
  2. select rung
  3. degradation receipt
Memory, control, and adaptation add depth without hiding their receipts.

Next

Open the proof surfaces before trusting the summary.

The feature set matters because it leaves an artifact trail. See what a completed run stores, then follow an illustrative route end to end.