Why This Matters

Steering techniques only stick if you apply them together. This checkpoint takes a vague prompt and evolves it through specificity, role, schema, and template — producing a verifiable, reusable prompt.

Concept Explained

Take a real vague prompt. Evolve it through: Level 0 (vague) → Level 1 (specific) → Level 2 (role + constraints) → Level 3 (schema) → Level 4 (template + verifiable). Measure output quality at each level.

How It Works

Find a vague prompt from your history. Apply techniques from IN-01 to IN-07. Document the evolution. Run the final prompt and verify output matches the schema.

Step-by-Step Tutorial

1. Find vague prompt

2. Level 1: add specificity

3. Level 2: add role + constraints

4. Level 3: add schema

5. Level 4: template + verify

6. Document

Real-World Example

A developer's vague prompt 'fix this' evolved to: 'Act as senior TS engineer (role). Fix the null-pointer issue (specificity). Don't use try/catch; use Result type (negative + positive). Output as unified diff (format). Template: fix-{{issue-type}}.md (template). Verify by running tests (verification).' Quality improved from generic to a clean, applicable diff in 5 iterations.

Example Prompts / Commands / Code

Level 0 (vague):
  "fix this"

Level 1 (specific):
  "Fix the TypeError when input is null in the formatDate function.
   The function should throw TypeError with a descriptive message."

Level 2 (role + constraints):
  "Act as a senior TypeScript engineer. Fix the TypeError when input is null.
   Don't use try/catch. Use the Result type from neverthrow.
   Output as unified diff."

Level 3 (schema):
  "Act as senior TS engineer. Fix the null TypeError.
   Output as JSON: { summary: string, diff: string, tests: string[] }
   No prose around the JSON."

Level 4 (template + verify):
  # templates/fix-null-issue.md
  Act as senior {{language}} engineer.
  Fix the {{error_type}} when {{condition}}.
  Don't use {{anti_pattern}}. Use {{preferred_pattern}}.
  Output as JSON: { summary, diff, tests }
  Verify: run {{test_command}}; all tests must pass.

Common Mistakes

• Skipping levels — evolution reveals what each technique contributes.
• Not measuring output quality at each level.
• Not converting the final prompt to a template — missing the reuse benefit.

Best Practices

• Apply techniques one at a time to isolate impact.
• Measure output quality at each level.
• Convert final prompt to a template for reuse.
• Build a verification step (test run, schema validation).
• Document the evolution as a case study.

Troubleshooting

Practical Exercise

Professional Challenge

Frequently Asked Questions

Further Reading

• Prompt engineering

Official References

• Prompt engineering

Related lessons: IN-01, IN-06