danmestas SPDD Iterative Review
Turn generated output into a controlled loop. Behavior first, code review second, prompt/code sync always.
Announce at start: "I'm using the spdd-iterative-review skill to classify feedback and keep the prompt and code synchronized."
Review order
- Behavior: Does the system satisfy acceptance criteria and safeguards?
- Prompt/code consistency: Does the code implement exactly what the canvas says?
- Architecture: Do layers, dependencies, interfaces, and responsibilities match Structure and Approach?
- Code quality: Check error handling, imports, dependency choices, magic values, long methods, duplication, and repository idioms.
- Test adequacy: Do tests cover normal, boundary, and error cases that can break?
Classify every finding
Logic correction
Observable behavior, business rule, data contract, security rule, performance guarantee, or acceptance criterion changes.
Action: run spdd-prompt-update first, then update code with spdd-generate.
Refactor
Internal structure changes with no observable behavior change.
Action: refactor code, prove behavior still passes, then run spdd-sync.
Prompt defect
The canvas is ambiguous, incomplete, or internally inconsistent.
Action: repair the canvas and re-review before code changes continue.
Code defect
The canvas is right, but code violates it.
Action: fix code under spdd-generate and rerun affected verification.
Output
Use this format:
## SPDD review
### Behavior evidence
- ...
### Findings
- [logic-correction|refactor|prompt-defect|code-defect] <finding> → <next skill/action>
### Prompt/code sync status
- Canvas describes current code: yes/no
- Required update: spdd-prompt-update / spdd-sync / none
Do not declare the slice complete while canvas and code disagree.