Write human-readable workflows that chain AI tools.
Install, configure, and sequence — without code.
## Step 1: Fetch Data ### Task Download all transactions from the payment API for last month. Handle pagination automatically. ### Flexibility [guided] Use any HTTP library. Caching is optional but recommended. ### Save as `/tmp/workflow/transactions.json` ### Success criteria - File exists and is valid JSON - Contains at least 100 records
The missing layer between rigid scripts and unpredictable AI agents
Describe what you want to achieve, not how to do it. Let the AI choose the best approach.
Mark exactly where AI can improvise: [strict], [guided], or [autonomous].
Steps communicate through files, not code. Clear inputs and outputs for each step.
Define fallbacks in plain language: "If X fails → try Y". No try/catch needed.
Verifiable conditions that confirm step completion. AI validates its own work.
The workflow IS the documentation. Readable by developers, managers, and AI alike.
Chain AI tools in declarative workflows
## Step 1: Setup Database Access ### Dependencies npm install -g @anthropic/mcp-server-postgres ### Task Verify connection and list available tables. ## Step 2: Query Data ### Task Using the Postgres MCP, fetch all orders from last month. Join with customers table to get contact info. ## Step 3: Generate Report ### Dependencies Read skill at `/mnt/skills/public/docx/SKILL.md` ### Task Following the docx skill instructions, create a professional Word document with sales summary.
Find the right balance between control and flexibility
| Approach | Flexibility | Predictability | Requires Code |
|---|---|---|---|
| Traditional Scripts | None | High | Yes |
| LangChain / DSPy | Medium | Medium | Yes (Python) |
| IntentFlow | Controlled | High | No |
| AutoGPT-style | Full | Low | No |
Write your first workflow in 5 minutes. No installation required.