AI coding tools
Windsurf Operating Card
Use Windsurf Cascade with scoped Code and Chat modes, durable rules, reviewed tools, and named checkpoint recovery.
- Format
- Reference
- Level
- Beginner
- Audience
- Practitioner, Developer
- Owner
- Project42 Editorial
- Review cadence
- Every 30 days
- Prerequisites
- A supported Windsurf installation; An approved account and model policy; A version-controlled workspace
Choose Chat or Code based on required authority
Cascade provides conversational and coding modes, tool calls, plans, todo lists, terminal awareness, and checkpoints. Use Chat when explanation or investigation is sufficient and Code only when edits or commands are authorized.
Generated Memories are local context; use versioned workspace Rules or AGENTS.md for durable team guidance that reviewers can inspect.
Make plans, rules, and checkpoints reviewable
State the goal, allowed files, tool boundary, and tests before implementation. Review plans, queued messages, terminal actions, and the complete diff rather than accepting a successful-looking summary.
Create named checkpoints around risky experiments and keep Git as the authoritative delivery record. Scope ignore files and repository rules according to data classification.
Task: [one observable result]
Scope: [workspace and paths]
Permissions: [Chat/Code, terminal, edit, tools, and network]
Rules: [AGENTS.md and workspace rules]
Checkpoint: [named recovery point]
Verification: [plan, diff, tests, and runtime result]
Recovery: [revert checkpoint, restore Git, reconcile side effects]Expected evidence and verification
Expected evidence includes the active rules, selected mode, reviewed plan and tool calls, named checkpoint, Git diff, and exact check output. Verify that local Memories did not become unreviewed project policy.
If Cascade loses context or crosses scope, stop queued work, preserve the trace, restore the named checkpoint or Git state, and retry with explicit durable rules and a smaller task.