Synthvora Growth
Coding

Windsurf Prompts

Agentic IDE by Codeium that pairs with developers to write and refactor code.

10 ready-to-use prompts

Researched for Windsurf real features. Replace the [BRACKETED] parts with your own details, copy, and paste.

1Ship a Feature With the Cascade Agent
Act as a full-stack developer. I need to add [FEATURE] to [PROJECT] ([STACK]), which will touch roughly [NUMBER] files across [FRONTEND/BACKEND]. Give me the Windsurf Cascade workflow: open the project, switch to Write mode, and prompt 'implement [FEATURE]: update [FILE 1] for the UI, [FILE 2] for the API, and [FILE 3] for the database migration; run [TEST COMMAND] after each change and fix failures before continuing'; review each diff Cascade proposes before accepting; use Checkpoints before the risky [STEP]. Include the approval discipline (never auto-approve [FILE TYPE]), what to do when Cascade loops on [ERROR] ([ESCALATION]), and the final verification (tests green, preview running, [MANUAL CHECK]).
Tip: Set a Checkpoint before the risky step; rolling back a Cascade flow beats untangling it.
2Run Parallel Agents on Git Worktrees (Wave 13)
Act as a solo developer with [NUMBER] features in flight on [PROJECT]. Windsurf Wave 13 supports parallel multi-agent sessions on isolated Git worktrees. Design the session: create worktrees for [FEATURE A] on branch [BRANCH A] and [FEATURE B] on branch [BRANCH B]; spawn one Cascade agent per worktree with scoped prompts ('implement [FEATURE A] touching only [DIRECTORY]'); monitor both in separate panes; merge each only after [TEST COMMAND] passes in its worktree. Include the rules that prevent cross-contamination (no shared [FILE]), what to do when both agents need [SHARED MODULE], and the merge order that minimizes conflicts.
Tip: Scope each agent to its own directory; parallel agents only stay clean with hard boundaries.
3Arena Mode: Pick the Best Model for Your Codebase
Act as a tech lead. We have access to [MODEL A], [MODEL B], and Windsurf's SWE-1.5, and we argue about which writes better [LANGUAGE] code for [PROJECT]. Run a blind Arena Mode comparison: pick [NUMBER] representative tasks ([TASK 1: bug fix], [TASK 2: new endpoint], [TASK 3: refactor]); run each model on each task without looking at which is which; score the outputs on correctness, style match to our [STYLE GUIDE], and test pass rate. Include the scoring rubric (1 to 5 per dimension), the minimum sample ([NUMBER] tasks) before concluding, and how to lock the winner in as the team default in Windsurf settings.
Tip: Score blind and score on your own tasks; benchmarks never match your codebase's quirks.
4Plan Mode Before a Big Refactor
Act as a senior engineer about to refactor [MODULE] in [REPO] ([LANGUAGE], [NUMBER] files affected). Use Windsurf's Plan Mode first: prompt 'decompose the refactor of [MODULE] into steps: for each step list the files changed, the risk (low/medium/high), and the test that proves it worked; do not write code yet'; review the plan and reorder the [NUMBER] high-risk steps to come with extra verification; only then switch to Write mode and execute step by step. Include the plan review checklist (missing steps, wrong order, untested edges like [EDGE CASE]) and the rule: if the plan changes mid-execution, return to Plan Mode before continuing.
Tip: Never skip the plan review; a bad plan executed fast is just fast damage.
5Memories and Rules for Team Consistency
Act as an engineering manager for [TEAM NAME] ([TEAM SIZE] devs, [STACK]). Windsurf Memories and Rules can encode our standards so the AI follows them automatically. Write the rule set: code style ([STYLE GUIDE], max line length [NUMBER]), language preferences ([LANGUAGE VERSION], prefer [PATTERN] over [ANTI-PATTERN]), project conventions ([NAMING], [DIRECTORY LAYOUT]), and the forbidden list ([FORBIDDEN 1], [FORBIDDEN 2]). Include how to store team-wide rules versus personal memories, the onboarding step where new hires import them, and the quarterly review that prunes stale rules (symptom: Cascade citing [OUTDATED RULE]).
Tip: Write rules as the team actually works, not as you wish it worked; the AI will expose the gap.
6One-Click MCP: Connect Your Tools
Act as a developer whose workflow lives across [TOOL 1], [TOOL 2], and [DATABASE]. Windsurf offers one-click MCP server setup. Walk through it: add the MCP servers for [TOOL 1] and [TOOL 2] from the Windsurf MCP panel (no manual JSON config); verify each connection with a test query ('list my open [ITEMS] in [TOOL 1]'); then build the killer workflow: 'Cascade, take the [TICKET] from [TOOL 1], find the related code, implement the fix, and update the ticket with the PR link.' Include the permission scoping (read-only for [TOOL], write allowed for [OTHER TOOL]) and the audit habit (weekly review of what the agents touched).
Tip: Start MCP tools read-only; grant write access only after a week of clean reads.
7Checkpoint Discipline for Risky Edits
Act as a developer doing a scary multi-file change in [PROJECT]: [CHANGE DESCRIPTION] across [NUMBER] files. Define the Windsurf Checkpoint discipline: create a checkpoint before each of the [NUMBER] phases ([PHASE 1], [PHASE 2], [PHASE 3]); name each checkpoint descriptively ([NAMING SCHEME]); if phase [N] breaks [TEST], roll back to the checkpoint instead of forward-fixing; document what was tried in the commit message when retrying. Include the decision rule for rollback versus repair (rollback if [CONDITION], repair if [OTHER CONDITION]) and how to explain the checkpoint history in the final PR.
Tip: Name checkpoints by phase and intent; 'checkpoint 3' tells future-you nothing.
8Terminal AI for Debugging Sessions
Act as a [STACK] developer staring at a failing [COMMAND] output: [ERROR SUMMARY]. Use Windsurf's Terminal AI (Ctrl+I): invoke it on the error and ask 'explain this failure and suggest the [NUMBER] most likely fixes for [PROJECT]'; apply the top suggestion, re-run, and iterate; when the fix works, ask 'what was the root cause in one paragraph, and how do I prevent it in [CI/CD]?' Include the escalation path (after [NUMBER] failed AI suggestions, switch to manual debugging of [AREA]), and the habit of saving the winning fix as a Memory so Cascade suggests it next time.
Tip: After the terminal AI fixes it, save the fix as a Memory; the second occurrence should be instant.
9Supercomplete for Boilerplate Speed
Act as a [LANGUAGE] developer writing [BOILERPLATE TYPE] across [PROJECT]. Windsurf's Supercomplete suggests entire functions from project context, not just lines. Optimize the workflow: prime it by opening the [NUMBER] most representative files ([FILE 1], [FILE 2]) so suggestions match our patterns; accept full-function completions for [REPETITIVE TASK] but type [COMPLEX LOGIC] manually; when a suggestion is close but wrong, edit [SPECIFIC PART] instead of rejecting outright (this trains your sense of its strengths). Include the weekly calibration: which [NUMBER] files to keep open as context anchors, and the signal that Supercomplete has learned our codebase (suggestions matching [TEAM PATTERN] unprompted).
Tip: Keep your best pattern files open as context anchors; Supercomplete mirrors what it sees.
10Migrate to Devin Desktop (Post-Rebrand Checklist)
Act as a team lead. Cognition rebranded Windsurf as Devin Desktop in June 2026, with Devin Local replacing the Cascade agent (legacy Cascade usable through July 1, 2026) and an Agent Command Center as the default surface. Build our migration checklist for [TEAM SIZE] developers: back up Windsurf settings, keybindings, and Rules; install Devin Desktop and verify the JetBrains plugin still carries the Windsurf name if the team uses JetBrains; re-test the [NUMBER] critical workflows ([WORKFLOW 1], [WORKFLOW 2]) under Devin Local; confirm ACP support still connects our third-party agents ([AGENT NAMES]). Include the rollback plan if Devin Local behaves differently on [PROJECT], and the comms message to the team explaining what changed and what did not.
Tip: Re-test your critical workflows under Devin Local before the July 1 legacy cutoff; do not discover differences mid-sprint.
Back to all 214 platforms