Prompt Chaining

The Problem

Single prompts hit a ceiling when tasks require multiple transformations — analyzing code, then generating tests, then writing documentation. Trying to do everything in one prompt produces shallow results across all steps. Prompt chaining breaks complex workflows into discrete stages, where each prompt does one thing well and passes its output to the next.

The Prompt

We are executing a multi-step workflow. I will give you one step at a time.

WORKFLOW OVERVIEW:
Step 1: [Brief description of step 1]
Step 2: [Brief description of step 2]
Step 3: [Brief description of step 3]

CURRENT STEP: Step 1
OBJECTIVE: [Detailed objective for this step]
INPUT: [Data or context for this step]

OUTPUT FORMAT:
[Exact format needed so the output can feed into Step 2]

When complete, end your response with:
---CHAIN OUTPUT---
[structured output for next step]
---END CHAIN---

Example Output

STEP 1 ANALYSIS COMPLETE:

Found 4 public API endpoints without authentication middleware:
- POST /api/comments (line 45, routes/comments.ts)
- GET /api/export (line 78, routes/export.ts)
- DELETE /api/cache (line 12, routes/admin.ts)
- PUT /api/settings (line 91, routes/settings.ts)

---CHAIN OUTPUT---
{
  "unprotected_endpoints": [
    {"method": "POST", "path": "/api/comments", "file": "routes/comments.ts", "line": 45},
    {"method": "GET", "path": "/api/export", "file": "routes/export.ts", "line": 78},
    {"method": "DELETE", "path": "/api/cache", "file": "routes/admin.ts", "line": 12},
    {"method": "PUT", "path": "/api/settings", "file": "routes/settings.ts", "line": 91}
  ]
}
---END CHAIN---

When to Use

Use prompt chaining for multi-phase tasks like code review pipelines, content creation workflows, data transformation sequences, or any process where quality improves when each step gets full attention. It is the foundation for building reliable AI-powered automation.

Pro Tips

• Define the chain output format upfront — structured data (JSON, CSV) chains more reliably than prose between steps.
• Keep each step focused — one task per prompt prevents quality degradation; resist the urge to combine steps.
• Include the workflow overview in every step — this helps the model understand what comes next and optimize its output accordingly.
• Add validation between steps — before feeding Step 1 output into Step 2, check it meets your expectations programmatically.