Cowork Plan-Before-Execute Pattern
CLAUDE.md pattern that instructs Cowork to always produce a written plan and get explicit approval before executing any multi-step task. Prevents irreversible actions from happening without human review.
- Require written plans before Claude executes multi-step tasks
- Add human review checkpoints to prevent irreversible mistakes
- Build accountability into every automated workflow
Cowork's agentic nature means it can take consequential actions (send emails, modify files, submit forms) before you realize what's happening. The plan-first pattern creates a mandatory review checkpoint for every non-trivial task.
Conservative users and client deployments where accountability and oversight of AI actions is required
https://x.com/HurynPawel/status/2028867432143429733
By Paweł Huryn
How to Get It
In Claude Cowork: Open the Customize panel in the sidebar → Folder Instructions, and paste the instructions below. Claude will follow them automatically in that folder.
In Claude Code: Add these instructions to your project's CLAUDE.md file, or paste them into your conversation and ask Claude to save them.
## Plan Before Execute Before executing any task with more than 2 steps: 1. Write a numbered plan with clear steps 2. Present the plan for my review 3. Wait for my approval before taking any action 4. Execute one step at a time, confirming after each Never skip the planning step. If I say "just do it", still present a brief plan first. If a step fails, stop and re-plan rather than pushing forward.
Trust Signals Personally Tested
Not yet scanned. Learn how to check this tool yourself.
Data & Access
Community Pulse Growing
Discussed on Reddit
- My shift supervisor at work was a petty tyrant who treated me like garbage. So I — Reddit · 8752 pts
- [Titanic] A Titanic video game spirals out of control in a frustrating 10 year s — Reddit · 3310 pts
- AIO for threatening to take my sister to court after her toddler destroyed my $2 — Reddit · 2479 pts
10 mentions across 1 sources
Reviewer notes
Personally Tested review. These are observations, not a security certification.
CLAUDE.md pattern only — no external dependencies. Critical safety pattern for client deployments. Pairs with delegation framework. Personally validated.
2026-03-29: CLAUDE.md snippet from Paweł Huryn's guide: 'Before executing any task with more than 2 steps: 1) Write a numbered plan. 2) List any files, emails, or external services that will be affected. 3) Ask me to confirm before proceeding. Exception: read-only operations (research, analysis, drafting) can proceed without confirmation.' This is the most important safety pattern for new Cowork deployments at clients. Pairs well with cowork-delegation-framework.
How to evaluate tools before deploying →
Data shown here comes from public APIs and automated scanning. Reviewer notes reflect one person's experience. This is not a security certification or legal recommendation. Always evaluate tools according to your own organization's policies.
Evaluation
CLAUDE.md pattern only — no external dependencies. Critical safety pattern for client deployments. Pairs with delegation framework. Personally validated.