Free printable workflow

Claude Code workflow checklist for safe, reviewable delivery.

Use these five gates before a significant coding task. Each gate asks for visible evidence, not confidence, so speed does not outrun scope, permissions, tests, or human review.

Reviewed by the ClaudeQuest team on .

Practise the first gate →

Five evidence gates

Do not move forward until the current gate has evidence.

GATE 01

Frame the task

Evidence outcome: A finish line that can be checked.

  • State one outcome and three observable acceptance criteria.
  • Name the files, services, or behaviors in scope.
  • Write down non-goals so an improvement does not become a rewrite.
  • Identify decisions that still require a human owner.

GATE 02

Ground the context

Evidence outcome: Repository truth before a proposed solution.

  • Read project instructions and the nearest relevant documentation.
  • Inspect related code, tests, interfaces, and recent patterns before editing.
  • Separate facts found in the repository from assumptions that need proof.
  • Exclude secrets, personal data, and unrelated files from the working context.

GATE 03

Set control boundaries

Evidence outcome: Useful autonomy without unreviewed consequences.

  • Start with inspection; approve mutation only after the target is resolved.
  • Use the narrowest filesystem, command, network, and tool permissions that work.
  • Require explicit approval for deletion, billing, credentials, production, or external messages.
  • Choose a recoverable branch, backup, or rollback path before consequential changes.

GATE 04

Implement in evidence-sized steps

Evidence outcome: A coherent change whose effect can be isolated.

  • Change one behavior or one falsifiable hypothesis at a time.
  • Keep the diff as small as possible while still completing the requested behavior.
  • Add or update the closest test when behavior changes.
  • Stop speculative fixes when the available evidence no longer supports them.

GATE 05

Verify and hand off

Evidence outcome: Reviewable evidence, remaining risk, and a clear next action.

  • Run focused tests first, then the broader quality checks appropriate to the risk.
  • Inspect the final diff for accidental scope, secrets, stale copy, and broken links.
  • Test the user-visible path in the environment where it will run.
  • Report what passed, what remains uncertain, and how the change can be rolled back.

Evidence ledger

Know what to keep—and when to stop.

Frame

Keep this evidence
Outcome, acceptance criteria, scope, non-goals
Stop when
The finish line cannot be observed

Ground

Keep this evidence
Relevant instructions, code paths, tests, interfaces
Stop when
The proposal depends on an unverified assumption

Control

Keep this evidence
Permission boundary and recovery path
Stop when
A consequential action lacks explicit authority

Implement

Keep this evidence
Small diff plus a testable hypothesis
Stop when
Several causes are changing at once

Verify

Keep this evidence
Test output, inspected diff, live-path check
Stop when
Success is asserted without evidence

Copy this task brief.

Outcome:
Acceptance criteria:
1.
2.
3.
In scope:
Out of scope:
Allowed actions:
Ask before:
Verification:
Remaining risk:

Escalate instead of guessing.

  • • The requested outcome conflicts with repository instructions.
  • • A deletion, credential, payment, production, or external message lacks authority.
  • • Tests cannot distinguish the proposed cause from an alternative.
  • • The change expands beyond the agreed files, behavior, or risk boundary.

Source and review basis

This independent checklist is informed by Anthropic’s Claude Code documentation on security, permissions, common workflows, and evidence-led development. ClaudeQuest turns those principles into an original five-gate learning aid; it is not official Anthropic material.

Primary-source ledger: Claude Code Security; Claude Code Best Practices; Claude Code Common Workflows. Reviewed August 4, 2026.

Turn the checklist into practice.

The free Claude Code module asks you to bound a real coding task, identify repository truth, and choose evidence before making a change.