Fix plan generation skipping TDD for config/tooling tasks

[chuanghiduoc]

Summary

Fixes #853 — Plans generated by writing-plans did not enforce TDD structure, especially when early tasks involved configuration or tooling. The plan reviewer also lacked TDD criteria, allowing non-TDD plans to pass review.

Root cause

Two gaps in the planning pipeline:

1. writing-plans/SKILL.md mentioned TDD only in passing ("DRY. YAGNI. TDD.") without enforcement. No <HARD-GATE>, no anti-rationalization guidance. When early tasks were config/tooling, the planner skipped TDD for those — and the habit carried to subsequent tasks.

2. plan-document-reviewer-prompt.md checked Completeness, Spec Alignment, Task Decomposition, and Buildability — but had no TDD criteria. Non-TDD plans were approved without question.

Changes

skills/writing-plans/SKILL.md:

• Added <HARD-GATE> requiring RED-GREEN-REFACTOR for every code-producing task
• Added concrete guidance for tasks that feel "untestable": config (smoke test), build pipeline (output check), scaffolding (startup check)
• Added anti-pattern section: "Tests at the end" — plans that defer all tests to a final task are explicitly flagged as NOT TDD

skills/writing-plans/plan-document-reviewer-prompt.md:

• Added "TDD Structure" category to the reviewer checklist
• Reviewer now checks: test before implementation in every task, no deferred test tasks

Test plan

• Brainstorm a project with config/tooling first steps → verify plan uses RED-GREEN per task
• Verify plan reviewer flags plans with implementation-before-test ordering
• Verify plan reviewer flags plans with a single "Add tests" task at the end

[Copilot]

Pull request overview

This PR tightens the “writing-plans” skill and its plan reviewer prompt to explicitly enforce TDD (RED/GREEN/REFACTOR) per task, especially preventing plans from deferring tests to the end when early tasks are configuration/tooling.

Changes:

• Added a <HARD-GATE> section to skills/writing-plans/SKILL.md requiring TDD structure for code-producing tasks, including guidance for “untestable” seeming tasks.
• Added an explicit anti-pattern warning against “tests at the end” plans.
• Updated skills/writing-plans/plan-document-reviewer-prompt.md to include “TDD Structure” in the reviewer checklist.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
skills/writing-plans/SKILL.md	Introduces mandatory TDD gating and guidance to prevent plans skipping tests during config/tooling tasks.
skills/writing-plans/plan-document-reviewer-prompt.md	Adds reviewer criteria to catch non-TDD task ordering and deferred-test plans.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[chuanghiduoc]

Thanks for the review! All 3 comments addressed in c44a6d6:

1. HARD-GATE now includes explicit refactor step: write failing test → verify it fails → write minimal implementation → verify it passes → refactor → commit

2. Narrowed scope: Changed from "produces or modifies code" to "introduces or changes behavior". Refactor-only tasks now explicitly rely on existing test suite rather than requiring a new failing test.

3. Reviewer prompt aligned: Now matches SKILL.md wording — RED-GREEN-REFACTOR with commit, and notes refactor-only tasks may use existing tests.