windsurf-core-workflow-a
Uses Cascade Write mode to enable autonomous, multi-file code editing and feature implementation.
Install
mkdir -p .claude/skills/windsurf-core-workflow-a && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/4846" && unzip -o skill.zip -d .claude/skills/windsurf-core-workflow-a && rm skill.zipInstalls to .claude/skills/windsurf-core-workflow-a
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Execute Windsurf''s primary workflow: Cascade Write mode for multi-fileKey capabilities
- →Create new files and modify existing code across multiple files
- →Run terminal commands and install dependencies
- →Debug errors by reading terminal output
- →Review and revert individual or all changes made by the agent
- →Iterate on errors by providing context for targeted fixes
- →Reference specific files or directories for precise agent actions
How it works
The agent creates and modifies files, runs terminal commands, installs dependencies, and debugs errors based on a natural language prompt. It shows its reasoning and plan before execution, allowing for review and iteration.
Inputs & outputs
When to use windsurf-core-workflow-a
- →Build features across multiple files
- →Perform cross-file refactoring
- →Automate dependency installation and code updates
About this skill
Windsurf Core Workflow A — Cascade Code Mode
Overview
Devin Desktop is the current name for Windsurf. Cascade Code mode is its agentic editing workflow: it can create and modify files, call tools, run approved terminal commands, inspect diagnostics, and iterate on failures.
Prerequisites
- Devin Desktop with Cascade enabled
.devin/rules/project.mdconfigured (seewindsurf-sdk-patterns)- Git initialized with clean working tree
Tool Use
- Use
Readto inspect only the repository files and configuration needed for the request. - Use
Grepto locate relevant settings, rules, logs, or code without broad collection. - Use
Writeonly for a new artifact the user requested; never write credentials or unreviewed production configuration. - Use
Editfor bounded, reviewable changes and preserve unrelated user work. - Use only the command-scoped
Bashentries declared in frontmatter, with non-destructive checks before mutations.
Instructions
Step 1: Create a Git Checkpoint
Always commit or stash before a Cascade session. Cascade writes directly to your files.
git add -A && git commit -m "checkpoint: before cascade session"
# Or for uncommitted work:
git stash push -m "pre-cascade stash"
Step 2: Open Cascade in Code Mode
Open the Cascade panel and select Code rather than Chat. Keyboard shortcuts can vary by imported profile, so use the visible command or confirm the current binding before documenting it.
- Create and modify files
- Run terminal commands (with Turbo or per-command approval)
- Install dependencies
- Read terminal output for debugging
- Open browser previews
Step 3: Write an Effective Prompt
Structure your prompt with scope, specifics, and constraints:
"In src/services/, create a NotificationService that:
1. Sends email via Resend API (already in package.json)
2. Sends Slack messages via webhook URL from env
3. Uses the Result<T,E> pattern from src/types/result.ts
4. Includes retry logic with exponential backoff (max 3 retries)
5. Add unit tests in tests/services/notification.test.ts
Don't modify any existing files except to add exports."
Step 4: Review Cascade's Plan and Execution
Cascade shows its reasoning and plan before executing:
Cascade output flow:
1. "I'll create the notification service with email and Slack support..."
2. Creates src/services/notification.ts (shows diff)
3. Creates tests/services/notification.test.ts (shows diff)
4. Runs: npm install resend (if needed)
5. Runs: npx vitest run tests/services/notification.test.ts
6. Reports results
Review each file diff in the Cascade output. Then:
- Revert individual steps by hovering over a step and clicking the revert arrow
- Revert all to return to the state before the Cascade session
- Create named checkpoints for complex multi-step sessions
Step 5: Iterate on Errors
If tests fail, Cascade retains context about what it just did:
"The test for sendSlack is failing with 'fetch is not defined'.
Fix it by using the node:test built-in fetch mock."
Cascade reads the error, understands its own recent changes, and applies targeted fixes.
Step 6: Use @ Mentions for Precision
@src/types/result.ts — force Cascade to read this file for the Result pattern
@src/services/ — reference the entire services directory for consistency
@web resend API docs — search the web for current Resend documentation
Output
- Multi-file code changes applied by Cascade
- Terminal commands executed (installs, tests, builds)
- Test results confirming implementation correctness
- Reviewable diffs for every file change
Error Handling
| Issue | Cause | Solution |
|---|---|---|
| Cascade modifies wrong files | Vague prompt | Specify exact file paths and constraints |
| Changes break existing tests | No constraint on existing code | Add "don't modify existing files" to prompt |
| Cascade loops on error | Insufficient context | Paste the full error message, reference relevant files |
| Code mode is unavailable | Account, policy, or quota restriction | Check the visible account and organization state |
Cascade ignores .devin/rules/project.md | Wrong path, trigger, or file over 12,000 characters | Validate frontmatter and split the workspace rule |
Examples
Full-Stack Feature
"Add a user profile page:
1. Create app/profile/page.tsx as a Server Component
2. Create app/profile/edit/page.tsx as a Client Component with form
3. Add GET /api/profile and PUT /api/profile route handlers
4. Use the existing UserSchema from lib/types/user.ts for validation
5. Style with Tailwind matching the existing design system
6. Add tests for both API routes"
Refactoring Task
"Extract the authentication logic from src/middleware/auth.ts into:
- src/services/auth.ts (JWT validation, token refresh)
- src/services/session.ts (session management)
Update all imports across the codebase. Run tests after."
Resources
Related Skill
Continue with windsurf-core-workflow-b to convert repeated session guidance into the appropriate Rules, Skills, Workflows, or Memories layer.
When not to use it
- →When the prompt is vague, leading to unintended file modifications
- →When existing tests should not be modified by the agent
- →When the agent is looping on an error due to insufficient context
Prerequisites
Limitations
- →Cascade may modify wrong files if the prompt is vague
- →Changes may break existing tests if no constraints are provided
- →Cascade may loop on errors if insufficient context is given
How it compares
This workflow automates multi-file code changes, dependency installation, and error debugging from a single prompt, unlike manual coding that requires separate actions for each task.
Compared to similar skills
windsurf-core-workflow-a side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| windsurf-core-workflow-a (this skill) | 1 | 2mo | Review | Intermediate |
| agent-factory | 8 | 11mo | Review | Advanced |
| claude-opus-4-5-migration | 9 | 10mo | No flags | Beginner |
| subagent-driven-development | 14 | 9mo | No flags | Advanced |
Try saying
Example prompts that trigger this skill in your AI assistant.
More by jeremylongshore
View all by jeremylongshore →You might also like
agent-factory
alirezarezvani
Claude Code agent generation system that creates custom agents and sub-agents with enhanced YAML frontmatter, tool access patterns, and MCP integration support following proven production patterns
claude-opus-4-5-migration
anthropics
Migrate prompts and code from Claude Sonnet 4.0, Sonnet 4.5, or Opus 4.1 to Opus 4.5. Use when the user wants to update their codebase, prompts, or API calls to use Opus 4.5. Handles model string updates and prompt adjustments for known Opus 4.5 behavioral differences. Does NOT migrate Haiku 4.5.
subagent-driven-development
davila7
Use when executing implementation plans with independent tasks in the current session
coding-agent
openclaw
Run Codex CLI, Claude Code, OpenCode, or Pi Coding Agent via background process for programmatic control.
continuous-learning-v2
affaan-m
Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents.
copilot-sdk
github
Build agentic applications with GitHub Copilot SDK. Use when embedding AI agents in apps, creating custom tools, implementing streaming responses, managing sessions, connecting to MCP servers, or creating custom agents. Triggers on Copilot SDK, GitHub SDK, agentic app, embed Copilot, programmable agent, MCP server, custom agent.