| name | specmint-tdd-html |
| description | TDD-first spec management for AI coding workflows, producing canonical SPEC.html files with red-green-refactor state. Use this companion when the user explicitly mentions specs, forging, structured planning, resume/status, implementing a spec, red-green-refactor, or running tests. Do NOT trigger on ordinary coding tasks that do not ask for spec management.
|
Spec Mint TDD HTML — Kluris Companion
Turn feature requests into HTML specs and implement them with strict TDD. This
HTML variant uses .specs/<id>/SPEC.html as the canonical spec document. The
HTML file carries metadata, acceptance criteria, phases, TEST/IMPL task pairs,
data-status progress, data-tdd-phase red-green-refactor state, diagrams,
code previews, the TDD log, decisions, and deviations.
When this file is referenced from a Kluris-generated brain skill, read and
follow it directly from ~/.kluris/companions/specmint-tdd-html/SKILL.md.
Do not ask the user to install Specmint separately. Kluris vendors companions
as single SKILL.md files, so every rule needed for companion mode is below.
Critical Invariants
- Canonical files
- Registry:
.specs/registry.md (markdown index)
- Per-spec document:
.specs/<id>/SPEC.html
- Research notes:
.specs/<id>/research-*.md
- Interview notes:
.specs/<id>/interview-*.md
- Scratch, if needed:
.specs/<id>/artifacts/
- Authority rule: the JSON inside
<script type="application/json" id="spec-meta">
is authoritative for identity and spec status. data-status attributes on
phases, tasks, and acceptance criteria are authoritative for progress.
data-tdd-phase on implementation tasks records red, green, or
refactor state when a cycle is in progress.
- TDD invariant: no production code without a failing test. Capture the
RED failure before writing implementation, make GREEN with the smallest
production change, then REFACTOR under green tests.
- HTML is the source of truth: do not create or update a parallel markdown
spec for an HTML companion spec.
- Progress tracking is sacred: after completing any task, immediately
update
SPEC.html (data-status, data-tdd-phase, phase transitions,
updated date, TDD log) and .specs/registry.md (progress/date). Re-read
both files before moving on.
- Single-file companion mode: use inline HTML/CSS/JS patterns in the
SPEC.html you write. Do not rely on unavailable sibling template or asset
files from a separate plugin install.
Registry Format
Create .specs/registry.md if missing:
# Spec Registry
| ID | Title | Status | Priority | Progress | Updated |
|----|-------|--------|----------|----------|---------|
| user-auth-system | User Auth System | active | high | 0/12 | 2026-05-23 |
The registry is a denormalized index only. If it conflicts with SPEC.html,
trust SPEC.html and repair the registry on the next write.
Session Start and Resume
If .specs/registry.md exists, check for an active row. When the user says
resume, status, or asks what was in progress:
- Read the active registry row.
- Load
.specs/<id>/SPEC.html.
- Count task progress from
<li class="task" data-status="...">.
- Current phase is the first
<details class="phase" data-status="in-progress">.
- Current task is the first pending task in that phase.
- If current task is an IMPL task, inspect
data-tdd-phase and the TDD Log.
- Present:
Resuming: <Title> (<id>)
Progress: <done>/<total> tasks
Phase: <phase title>
Current: <task code and text>
TDD phase: <RED/GREEN/REFACTOR or next TEST task>
Begin work on the current task unless the user asked only for a status report.
Forging a TDD HTML Spec
The forge workflow produces .specs/ files only. It does not implement
application code.
- Generate a lowercase hyphenated spec ID.
- Collision-check
.specs/<id>/SPEC.html and the registry.
- Create
.specs/<id>/ and .specs/registry.md if needed.
- Research deeply:
- map relevant code paths, dependency flow, and current test patterns;
- inspect fixtures, existing integration tests, and service boundaries;
- search current docs/best practices for important libraries;
- identify external services that must be mocked or replaced by local test
fixtures;
- identify risks, edge cases, and open questions.
- Save research to
.specs/<id>/research-01.md.
- Interview the user with targeted questions about behavior, acceptance
criteria, and test boundaries. Save answers to
.specs/<id>/interview-01.md.
- Repeat research/interview if important choices remain unclear.
- Write
.specs/<id>/SPEC.html from the self-contained HTML format below.
- Update the registry row to
active and 0/N progress.
- Validate the HTML/JSON structure, present the spec, and wait for explicit
approval before implementation.
SPEC.html Format
Use this structure. Keep region comments; they make future edits reliable.
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Spec: User Auth System</title>
<script type="application/json" id="spec-meta">{"id":"user-auth-system","title":"User Auth System","status":"active","created":"2026-05-23","updated":"2026-05-23","priority":"high","tags":["auth"],"mockup-fidelity":"hi-fi"}</script>
<style>
body{font-family:Inter,system-ui,sans-serif;margin:0;background:#0f172a;color:#e5e7eb;line-height:1.55}
main{max-width:1100px;margin:0 auto;padding:32px}.card{background:#111827;border:1px solid #334155;border-radius:16px;padding:20px;margin:16px 0}
.pill{display:inline-block;border-radius:999px;padding:2px 10px;background:#334155}.pill--completed{background:#065f46}.pill--in-progress{background:#1d4ed8}.pill--pending{background:#475569}.pill--blocked{background:#7f1d1d}.pill--red{background:#7f1d1d}.pill--green{background:#065f46}.pill--refactor{background:#854d0e}
.task,.ac-item{margin:10px 0;padding:10px;border-left:4px solid #475569;background:#0b1220}.task[data-status="completed"],.ac-item[data-status="completed"]{border-color:#10b981}.task[data-status="in-progress"],.phase[data-status="in-progress"]{border-color:#60a5fa}.task[data-status="blocked"],.phase[data-status="blocked"]{border-color:#f87171}
pre{overflow:auto;background:#020617;border-radius:12px;padding:16px}.mockup{background:#f8fafc;color:#0f172a;border-radius:16px;padding:20px}.table{width:100%;border-collapse:collapse}.table td,.table th{border:1px solid #334155;padding:8px;text-align:left}
</style>
</head>
<body>
<main>
<header class="card">
<h1>User Auth System</h1>
<p><span class="pill pill--in-progress">Active</span> <span class="pill">high</span></p>
<dl><dt>Created</dt><dd>2026-05-23</dd><dt>Updated</dt><dd>2026-05-23</dd></dl>
</header>
<section class="card"><h2>Overview</h2><p>What is being built and why.</p></section>
<section class="card"><h2>Acceptance Criteria</h2><ul>
<li class="ac-item" data-status="pending">User can sign in with GitHub.</li>
</ul></section>
<section class="card"><h2>Architecture</h2><pre class="mermaid">sequenceDiagram
participant U as "User"
participant FE as "Frontend"
participant API as "Backend API"
U->>FE: "click sign in"
FE->>API: "POST /auth/github"
API-->>FE: "session response"
</pre></section>
<section class="card"><h2>Testing Architecture</h2><p>Use the project test runner. Mock only external network boundaries. Prefer real parsers, serializers, and storage layers where practical.</p></section>
<section class="card"><h2>Phases & Tasks</h2>
<details class="phase" open data-status="in-progress"><summary><strong>Phase 1: Auth Contract</strong> <span class="pill pill--in-progress">In progress</span></summary><ul class="task-list">
<li class="task" data-status="pending"><span class="task-code">TEST-AUTH-01</span> Write failing tests for GitHub callback validation.</li>
<li class="task" data-status="pending" data-tdd-phase="red"><span class="task-code">IMPL-AUTH-02</span> Implement callback validation until tests pass, then refactor.</li>
</ul></details>
</section>
<section class="card"><h2>Code Previews</h2><figure class="code-diff"><figcaption>Canonical test pattern</figcaption><pre><code>+ def test_github_callback_rejects_missing_code(): ...</code></pre></figure></section>
<section class="card"><h2>UI Mockups</h2><figure class="mockup"><h3>Sign in</h3><button>Continue with GitHub</button></figure></section>
<section class="card"><h2>TDD Log</h2><table class="table"><thead><tr><th>Task</th><th>RED</th><th>GREEN</th><th>REFACTOR</th></tr></thead><tbody></tbody></table></section>
<section class="card"><h2>Decision Log</h2><table class="table"><thead><tr><th>Date</th><th>Decision</th><th>Rationale</th></tr></thead><tbody></tbody></table></section>
<section class="card"><h2>Deviations</h2><table class="table"><thead><tr><th>Task</th><th>Spec Said</th><th>Actually Did</th><th>Why</th></tr></thead><tbody></tbody></table></section>
</main>
</body>
</html>
Authoring Rules
- Metadata JSON must be valid and single-source-of-truth for
id, title,
status, created, updated, priority, tags, and mockup-fidelity.
- Valid lifecycle values:
pending, in-progress, completed, blocked.
- TEST tasks must precede their matching IMPL tasks.
- Task codes use
TEST-<PREFIX>-NN and IMPL-<PREFIX>-NN.
- Every IMPL task must have a corresponding TDD Log row before it is marked
completed.
- Every non-trivial spec needs at least one Mermaid diagram and at least one
code preview showing a meaningful test, contract, schema, API, or algorithm.
- Quote Mermaid participant aliases and message labels every time.
- Never leave
TBD, TODO, placeholder, figure out, or unresolved choices
in the final spec. Ask the user instead.
TDD Implementation Flow
When the user asks to implement:
- Read the active registry row and load
SPEC.html.
- Select scope: current task, a named phase, or all remaining phases.
- For each TEST task:
- write the failing test;
- run the exact test command;
- verify the failure is for the expected reason;
- mark the TEST task completed only after the RED failure is captured.
- For each IMPL task:
- start with
data-tdd-phase="red" if not already present;
- write the smallest production change to pass the paired test;
- run the exact test command and capture GREEN;
- refactor only while tests remain green;
- append/update the TDD Log row;
- mark the IMPL task completed.
- After each task:
- update the
updated date in spec-meta and the visible header;
- recompute progress directly from task
data-status counts;
- update
.specs/registry.md;
- re-read both files to verify progress is in sync.
- Log durable decisions in the Decision Log.
- Log implementation drift in Deviations.
- Before claiming phase/spec completion, show actual test command output.
Blocking Rule
If the next required test cannot be written or made to fail for the expected
reason, stop and ask. Do not write production code to bypass the RED phase.
Pausing and Completion
Pause by setting spec status to paused in spec-meta, changing the visible
status pill, updating the registry, and recording durable context.
Complete only when all tasks and acceptance criteria are complete, all TDD Log
rows have RED/GREEN/REFACTOR evidence, all phases are complete, the full
relevant test suite has been run, and the registry row matches the derived task
progress from SPEC.html.