-
Notifications
You must be signed in to change notification settings - Fork 2
Closed
Labels
enhancementNew feature or requestNew feature or request
Description
Summary
Add two new templates that produce structured prompts for code generation and test generation — designed to pair with audit-code-compliance and audit-test-compliance in a generate/verify loop.
The Generate/Verify Pattern
PromptKit assembles two prompts from the same requirements spec:
- Generation prompt: "Build code/tests that implement this spec"
- Audit prompt: "Verify the code/tests implement this spec"
The engineer feeds the generation prompt to Copilot/Claude to produce code, then feeds the audit prompt to verify it. Same spec drives both, so they can't drift. Fix findings, re-audit, repeat until clean.
New Templates
author-implementation-prompt
- Input: requirements-document + design-document + language/framework constraints
- Output: A structured prompt document (not code) that a coding agent consumes to generate a spec-compliant implementation
- Pairs with:
audit-code-compliance(same requirements input)
author-test-prompt
- Input: requirements-document + validation-plan + test framework conventions
- Output: A structured prompt document (not code) that a coding agent consumes to generate spec-compliant tests
- Pairs with:
audit-test-compliance(same requirements + validation plan input)
New Personas
implementation-engineer
- Stance: "Build correct, maintainable code from a specification"
- Different from
systems-engineer(which debugs/investigates) — this one creates - Behavioral constraints: spec traceability in code (comments referencing REQ-IDs), defensive error handling, constraint enforcement, no undocumented behavior
test-engineer
- Stance: "Write thorough tests that verify a specification"
- Different from
specification-analyst(which audits tests) — this one writes them - Behavioral constraints: coverage strategies, negative/boundary cases, acceptance criteria mapping, assertion quality over test quantity
Key design principle
PromptKit does not write code. It engineers the prompts that bracket the code — one to generate, one to verify. The output is always a document.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
enhancementNew feature or requestNew feature or request