make-spec — HTML design-doc scaffold (P18 Category-C)
One template + one theme + one filename convention — so the writer
spends 100% of energy on substance, 0% on boilerplate.
Why this skill exists
Since P18's promotion 2026-05-13, the workspace produces substantive
design docs as native HTML rather than markdown:
| Type |
Location |
Examples (last 7d) |
| Spec |
docs/specs/ |
2026-05-22-broker-selection-cross-asset.html, 2026-05-23-life-houston-runtime-integration.html |
| Plan |
docs/plans/ |
2026-05-22-feature-flag-design.html, 2026-05-22-houston-advanced-settings.html, 2026-05-22-houston-developer-mode.html |
| Report |
docs/reports/ |
2026-05-22-houston-dogfood-pattern.html, 2026-05-22-mission-control-vs-houston.html |
| ADR |
docs/adrs/ |
(folder reserved; pattern proven via broker-selection ADR in docs/specs/) |
| PR explainer |
docs/pr-explainers/PR-<n>.html |
(P18-reserved path) |
All 7 hand-rolled HTMLs above share the same 70-line CSS preamble:
--bg #0e1116, --ink #e7ecf2, --accent #7ec4ff, --accent-2 #b58cff,
h1-h4 with identical sizes/margins/colors, table with hover, callout
classes (warn/bad/ok/big), tag chips (info/warn/ok/bad/pick/cfd),
TOC block, footnote/cite styling.
Before this skill, every HTML doc rebuilt that boilerplate from
memory or copy-paste. The boilerplate IS the Broomva visual identity for
internal docs; centralizing it makes the visual identity stable AND
saves ~5 minutes per doc.
What this skill provides
references/theme.css — the canonical 70-line stylesheet,
verbatim from the proven specs. Reference, don't recopy.
references/template-spec.html — the one shipped base
template (the spec layout). Every variant below is generated
from this base, not loaded from its own file: start from
template-spec.html, then swap the title prefix, TOC sections,
and section skeleton per the variant's section list. No separate
template-plan/adr/report/pr-explainer.html files ship — the
variants are prose skeletons applied to the spec base.
The four variants (generated, not shipped — apply to the
template-spec.html base + the canonical skeleton below):
The base template has no Non-goals or Alternatives-considered slot, and
spec-contract requires both on the spec profile — so the shipped scaffold
fails the content gate it now composes with. Add both sections when generating
from it; they are the two classes the workspace corpus is weakest on (7.6%
each), which is exactly why a scaffold that omits them perpetuates the gap.
- Plan variant — base + sub-phase table + acceptance-criteria
sections.
- ADR variant — base + Status / Context / Decision /
Consequences / Alternatives sections — the
adr profile in
spec-contract, which enforces them rather than listing them.
- Report variant — base + Executive summary / Findings /
Recommendations / Appendix sections.
- PR-explainer variant — base + What changed / Why / Test plan
/ Risk / Rollout sections.
When to invoke
- Substantive design doc (>100 lines OR contains ≥1 of: decision
matrix, multi-row PR table, regulatory citations, multi-stage plan).
- ADR for any architectural decision worth keeping for posterity.
- PR explainer for a substantive PR (>200 LOC OR public API OR
governance-class change — same trigger as P20's review requirement).
- Plan for any multi-week / multi-PR arc that a stakeholder will
review (not just the executor).
- Report synthesizing prior research, briefings, or post-mortems.
Carve-outs (do not invoke)
- Brief docs <100 lines → markdown is fine (P18 audience rule: the
HTML payoff doesn't justify the 2-4× generation cost on short docs).
- Agent-loaded references (SKILL.md, entity pages, AGENTS.md) →
always markdown (Category A: substrate stays markdown).
- README / CHANGELOG / docs that GitHub renders → markdown (GitHub's
markdown renderer is the audience layer).
- PR descriptions (
gh pr create --body) → markdown (GitHub renders).
- Throw-away interactive UI → use Webdesign / impeccable / arcan-glass
skills; this skill is for static design docs only.
The canonical structure
Every variant follows the same skeleton:
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<title><Doc type> — <Doc title></title>
<meta name="viewport" content="width=device-width,initial-scale=1">
<style>
/* paste references/theme.css verbatim */
</style>
</head>
<body>
<h1><Doc title></h1>
<p class="subtitle"><One-line summary></p>
<p class="meta">
Author: <agent | user> · Generated: YYYY-MM-DD · Status: <draft | accepted | superseded><br>
Decision-class: <span class="tag info">substantive</span> · risk-class: <span class="tag warn">…</span><br>
Upstream context: <code>…</code>
</p>
<div class="callout warn">
<strong>Verification scope</strong>. <When this doc requires
out-of-band verification before action — e.g. CPA review, legal
review, deploy approval — state it here.>
</div>
<div class="toc">
<strong>Contents</strong>
<ol>
<li><a href="#section-1">…</a></li>
</ol>
</div>
<h2 id="section-1">1. <Section title></h2>
<!-- … -->
</body>
</html>
CSS theme (load from references/theme.css)
The theme is not open for tweaking inside individual docs. The
visual identity is stable across all internal Broomva HTMLs:
- Background:
#0e1116 (the deep ink that lets the --accent
blue and --accent-2 purple read cleanly)
- Ink:
#e7ecf2 primary, #98a2b3 dim, #6b7280 muted
- Accents:
#7ec4ff (h3 + tag.info) and #b58cff (tag.pick, the
"this is the decision" highlight)
- Status colors:
#66d699 ok, #ffd166 warn, #ff7a7a bad
- Code:
#0b0f15 background; monospaced via system stack
(ui-monospace, SFMono-Regular, "JetBrains Mono", Menlo)
- Sans:
-apple-system, "SF Pro Text", system-ui, sans-serif
To change the theme: amend references/theme.css and the change
propagates to every new doc. Existing docs are not retro-updated
(P13 stability budget — don't rewrite history for cosmetic deltas).
Tag system
The <span class="tag X">…</span> chips are the doc's status
vocabulary. Always-supported classes:
| Class |
Use |
Color |
info |
neutral metadata / classification |
--accent blue |
ok |
confirmed / passing / accepted |
--ok green |
warn |
needs verification / blocked-on-external |
--warn yellow |
bad |
rejected / failing / unsafe |
--bad red |
pick |
THE decision (uniquely highlighted) |
--accent-2 purple, bold |
cfd |
specific anti-pattern marker (synthetic exposure / not-real-ownership; from the broker-selection ADR vocabulary) |
--bad muted |
Callout system
Four callout flavors:
<div class="callout">…</div> <!-- neutral; accent border -->
<div class="callout ok">…</div> <!-- success / confirmed -->
<div class="callout warn">…</div> <!-- needs verification -->
<div class="callout bad">…</div> <!-- rejected / unsafe -->
<div class="callout big">…</div> <!-- the headline decision / TL;DR -->
File placement
| Doc type |
Path |
Filename |
| Spec |
docs/specs/ |
YYYY-MM-DD-<slug>.html |
| Plan |
docs/plans/ |
YYYY-MM-DD-<slug>.html |
| ADR |
docs/adrs/ (or docs/specs/ until the dir is canonical) |
YYYY-MM-DD-adr-<slug>.html |
| Report |
docs/reports/ |
YYYY-MM-DD-<slug>.html |
| PR explainer |
docs/pr-explainers/ |
PR-<n>.html (no date — PR number is the identifier) |
Slug names the topic, not the date. The date is the mtime.
The five anti-patterns this skill exists to prevent
| Anti-pattern |
Failure mode |
Fix |
| Rebuilding the theme inline |
70-line :root drift across docs; visual identity erodes. |
Always reference references/theme.css; don't paraphrase. |
| Markdown for a 200-line spec |
Tables don't render the way the writer pictured; ASCII pseudo-diagrams; reader bounces by line 80 (the trq212 ceiling). |
Apply the audience test: human-read substantive → HTML. |
| HTML for a 30-line note |
2-4× generation cost for no information-density payoff. |
Brief docs stay markdown. |
Title in <h1> differs from <title> |
Tab-bar text doesn't match doc heading; reader gets confused when 5 tabs open. |
Keep <title> and <h1> in sync; <title> adds the doc-type prefix ("ADR — …"). |
No meta line |
Reader can't tell author / date / status / risk-class without skimming. |
The <p class="meta"> line is mandatory: author + date + status + risk + upstream context. |
Composition rules
| Compose with |
When |
spec-contract |
Always, for a design doc. This skill owns how the doc LOOKS; spec-contract owns what is IN it (reversal-cost inclusion rule, required sections by class, ≥2 alternatives with rejection reasons, measurable acceptance, implementation-manual detector) plus the judgment rubric a script cannot run. Scaffold here, then gate with python3 skills/governance/spec-contract/scripts/spec_check.py <doc>. |
bookkeeping render |
NEVER for native HTML — bookkeeping render is for Category-B (MD canonical → HTML projection). This skill produces Category-C natives. They're disjoint per P18. |
handoff |
A handoff is markdown (agent-loaded) but may link to a make-spec HTML companion when the arc warrants. Handoff stays MD; companion is HTML. |
autonomous |
When /autonomous is mid-arc and a substantive plan emerges, fork the plan into docs/plans/<slug>.html via this skill, then continue execution. |
Webdesign / impeccable / arcan-glass |
Disjoint — those are for product-surface UI; this is for internal design docs. |
Validation (spec self-test)
These checks are about the shell, not the contents. Every item below can
pass on a document that argues nothing — that is not a defect in the list, it
is its scope. The contents are gated by spec-contract, whose checker is the
last item here and is the only one that reads what the doc says.
References
- Canonical theme source:
docs/specs/2026-05-22-broker-selection-cross-asset.html lines 8–71 (extracted into references/theme.css)
- P18 audience rule:
CLAUDE.md §P18 + AGENTS.md §P18
- trq212 thesis: The Unreasonable Effectiveness of HTML (May 2026, Claude Code team) — the empirical case behind P18
- Related skills:
bookkeeping (Category-B render), handoff (companion MD artifact), Webdesign (product UI, disjoint)
1---2name: make-spec3category: tooling4description: |5 Scaffold a substantive human-readable design doc (spec / plan / ADR /6 report / PR explainer) as native HTML using the workspace's canonical7 Broomva dark theme. Implements P18 (Format-Follows-Audience) for8 Category-C native artifacts — distinct from `bookkeeping render`,9 which projects Category-B markdown canonicals to HTML.10 Use when: (1) drafting a substantive design doc a human will actually11 read (>100 lines OR contains tables/diagrams/decision matrices),12 (2) writing an ADR (architectural decision record), (3) producing a13 plan a non-agent stakeholder will review, (4) writing a PR explainer14 for a substantive PR, (5) producing a report that synthesizes prior15 research. The skill ships theme.css + the spec HTML template, and16 generates the plan / adr / report / pr-explainer variants from that17 same theme + the prose skeletons below — so the agent doesn't18 rebuild the 70-line :root + h1-h4 + table + callout boilerplate19 every time.20 Triggers on "spec", "plan", "ADR", "decision record", "design doc",21 "explainer", "report", "html spec", "html doc", "Broomva html",22 "dark theme spec".23metadata:24 source: https://skillmd.com/skills/broomva/make-spec25---2627# make-spec — HTML design-doc scaffold (P18 Category-C)2829**One template + one theme + one filename convention — so the writer30spends 100% of energy on substance, 0% on boilerplate.**3132## Why this skill exists3334Since P18's promotion 2026-05-13, the workspace produces substantive35design docs as native HTML rather than markdown:3637| Type | Location | Examples (last 7d) |38|---|---|---|39| Spec | `docs/specs/` | `2026-05-22-broker-selection-cross-asset.html`, `2026-05-23-life-houston-runtime-integration.html` |40| Plan | `docs/plans/` | `2026-05-22-feature-flag-design.html`, `2026-05-22-houston-advanced-settings.html`, `2026-05-22-houston-developer-mode.html` |41| Report | `docs/reports/` | `2026-05-22-houston-dogfood-pattern.html`, `2026-05-22-mission-control-vs-houston.html` |42| ADR | `docs/adrs/` | (folder reserved; pattern proven via broker-selection ADR in `docs/specs/`) |43| PR explainer | `docs/pr-explainers/PR-<n>.html` | (P18-reserved path) |4445All 7 hand-rolled HTMLs above share the same 70-line CSS preamble:46`--bg #0e1116`, `--ink #e7ecf2`, `--accent #7ec4ff`, `--accent-2 #b58cff`,47h1-h4 with identical sizes/margins/colors, table with hover, callout48classes (warn/bad/ok/big), tag chips (info/warn/ok/bad/pick/cfd),49TOC block, footnote/cite styling.5051Before this skill, every HTML doc rebuilt that boilerplate from52memory or copy-paste. The boilerplate IS the Broomva visual identity for53internal docs; centralizing it makes the visual identity stable AND54saves ~5 minutes per doc.5556## What this skill provides57581. **`references/theme.css`** — the canonical 70-line stylesheet,59 verbatim from the proven specs. Reference, don't recopy.602. **`references/template-spec.html`** — the one shipped base61 template (the spec layout). Every variant below is **generated62 from this base**, not loaded from its own file: start from63 `template-spec.html`, then swap the title prefix, TOC sections,64 and section skeleton per the variant's section list. No separate65 `template-plan/adr/report/pr-explainer.html` files ship — the66 variants are prose skeletons applied to the spec base.6768The four variants (generated, not shipped — apply to the69`template-spec.html` base + the canonical skeleton below):7071> **The base template has no Non-goals or Alternatives-considered slot**, and72> `spec-contract` requires both on the `spec` profile — so the shipped scaffold73> fails the content gate it now composes with. Add both sections when generating74> from it; they are the two classes the workspace corpus is weakest on (7.6%75> each), which is exactly why a scaffold that omits them perpetuates the gap.76773. **Plan variant** — base + sub-phase table + acceptance-criteria78 sections.794. **ADR variant** — base + Status / Context / Decision /80 Consequences / Alternatives sections — the `adr` profile in81 `spec-contract`, which enforces them rather than listing them.825. **Report variant** — base + Executive summary / Findings /83 Recommendations / Appendix sections.846. **PR-explainer variant** — base + What changed / Why / Test plan85 / Risk / Rollout sections.8687## When to invoke8889- **Substantive design doc** (>100 lines OR contains ≥1 of: decision90 matrix, multi-row PR table, regulatory citations, multi-stage plan).91- **ADR** for any architectural decision worth keeping for posterity.92- **PR explainer** for a substantive PR (>200 LOC OR public API OR93 governance-class change — same trigger as P20's review requirement).94- **Plan** for any multi-week / multi-PR arc that a stakeholder will95 review (not just the executor).96- **Report** synthesizing prior research, briefings, or post-mortems.9798## Carve-outs (do not invoke)99100- Brief docs <100 lines → markdown is fine (P18 audience rule: the101 HTML payoff doesn't justify the 2-4× generation cost on short docs).102- Agent-loaded references (SKILL.md, entity pages, AGENTS.md) →103 always markdown (Category A: substrate stays markdown).104- README / CHANGELOG / docs that GitHub renders → markdown (GitHub's105 markdown renderer is the audience layer).106- PR descriptions (`gh pr create --body`) → markdown (GitHub renders).107- Throw-away interactive UI → use Webdesign / impeccable / arcan-glass108 skills; this skill is for static design docs only.109110## The canonical structure111112Every variant follows the same skeleton:113114```html115<!doctype html>116<html lang="en">117<head>118 <meta charset="utf-8">119 <title><Doc type> — <Doc title></title>120 <meta name="viewport" content="width=device-width,initial-scale=1">121 <style>122 /* paste references/theme.css verbatim */123 </style>124</head>125<body>126 <h1><Doc title></h1>127 <p class="subtitle"><One-line summary></p>128 <p class="meta">129 Author: <agent | user> · Generated: YYYY-MM-DD · Status: <draft | accepted | superseded><br>130 Decision-class: <span class="tag info">substantive</span> · risk-class: <span class="tag warn">…</span><br>131 Upstream context: <code>…</code>132 </p>133134 <div class="callout warn">135 <strong>Verification scope</strong>. <When this doc requires136 out-of-band verification before action — e.g. CPA review, legal137 review, deploy approval — state it here.>138 </div>139140 <div class="toc">141 <strong>Contents</strong>142 <ol>143 <li><a href="#section-1">…</a></li>144 </ol>145 </div>146147 <h2 id="section-1">1. <Section title></h2>148 <!-- … -->149</body>150</html>151```152153## CSS theme (load from `references/theme.css`)154155The theme is **not** open for tweaking inside individual docs. The156visual identity is stable across all internal Broomva HTMLs:157158- **Background**: `#0e1116` (the deep ink that lets the `--accent`159 blue and `--accent-2` purple read cleanly)160- **Ink**: `#e7ecf2` primary, `#98a2b3` dim, `#6b7280` muted161- **Accents**: `#7ec4ff` (h3 + tag.info) and `#b58cff` (tag.pick, the162 "this is the decision" highlight)163- **Status colors**: `#66d699` ok, `#ffd166` warn, `#ff7a7a` bad164- **Code**: `#0b0f15` background; monospaced via system stack165 (`ui-monospace, SFMono-Regular, "JetBrains Mono", Menlo`)166- **Sans**: `-apple-system, "SF Pro Text", system-ui, sans-serif`167168To change the theme: amend `references/theme.css` and the change169propagates to every new doc. Existing docs are not retro-updated170(P13 stability budget — don't rewrite history for cosmetic deltas).171172## Tag system173174The `<span class="tag X">…</span>` chips are the doc's status175vocabulary. Always-supported classes:176177| Class | Use | Color |178|---|---|---|179| `info` | neutral metadata / classification | `--accent` blue |180| `ok` | confirmed / passing / accepted | `--ok` green |181| `warn` | needs verification / blocked-on-external | `--warn` yellow |182| `bad` | rejected / failing / unsafe | `--bad` red |183| `pick` | THE decision (uniquely highlighted) | `--accent-2` purple, bold |184| `cfd` | specific anti-pattern marker (synthetic exposure / not-real-ownership; from the broker-selection ADR vocabulary) | `--bad` muted |185186## Callout system187188Four callout flavors:189190```html191<div class="callout">…</div> <!-- neutral; accent border -->192<div class="callout ok">…</div> <!-- success / confirmed -->193<div class="callout warn">…</div> <!-- needs verification -->194<div class="callout bad">…</div> <!-- rejected / unsafe -->195<div class="callout big">…</div> <!-- the headline decision / TL;DR -->196```197198## File placement199200| Doc type | Path | Filename |201|---|---|---|202| Spec | `docs/specs/` | `YYYY-MM-DD-<slug>.html` |203| Plan | `docs/plans/` | `YYYY-MM-DD-<slug>.html` |204| ADR | `docs/adrs/` (or `docs/specs/` until the dir is canonical) | `YYYY-MM-DD-adr-<slug>.html` |205| Report | `docs/reports/` | `YYYY-MM-DD-<slug>.html` |206| PR explainer | `docs/pr-explainers/` | `PR-<n>.html` (no date — PR number is the identifier) |207208Slug names the **topic**, not the date. The date is the mtime.209210## The five anti-patterns this skill exists to prevent211212| Anti-pattern | Failure mode | Fix |213|---|---|---|214| **Rebuilding the theme inline** | 70-line :root drift across docs; visual identity erodes. | Always reference `references/theme.css`; don't paraphrase. |215| **Markdown for a 200-line spec** | Tables don't render the way the writer pictured; ASCII pseudo-diagrams; reader bounces by line 80 (the trq212 ceiling). | Apply the audience test: human-read substantive → HTML. |216| **HTML for a 30-line note** | 2-4× generation cost for no information-density payoff. | Brief docs stay markdown. |217| **Title in `<h1>` differs from `<title>`** | Tab-bar text doesn't match doc heading; reader gets confused when 5 tabs open. | Keep `<title>` and `<h1>` in sync; `<title>` adds the doc-type prefix (`"ADR — …"`). |218| **No `meta` line** | Reader can't tell author / date / status / risk-class without skimming. | The `<p class="meta">` line is mandatory: author + date + status + risk + upstream context. |219220## Composition rules221222| Compose with | When |223|---|---|224| **`spec-contract`** | **Always, for a design doc.** This skill owns how the doc LOOKS; `spec-contract` owns what is IN it (reversal-cost inclusion rule, required sections by class, ≥2 alternatives with rejection reasons, measurable acceptance, implementation-manual detector) plus the judgment rubric a script cannot run. Scaffold here, then gate with `python3 skills/governance/spec-contract/scripts/spec_check.py <doc>`. |225| **`bookkeeping render`** | NEVER for native HTML — `bookkeeping render` is for Category-B (MD canonical → HTML projection). This skill produces Category-C natives. They're disjoint per P18. |226| **`handoff`** | A handoff is markdown (agent-loaded) but may *link to* a make-spec HTML companion when the arc warrants. Handoff stays MD; companion is HTML. |227| **`autonomous`** | When `/autonomous` is mid-arc and a substantive plan emerges, fork the plan into `docs/plans/<slug>.html` via this skill, then continue execution. |228| **`Webdesign` / `impeccable` / `arcan-glass`** | Disjoint — those are for product-surface UI; this is for internal design docs. |229230## Validation (spec self-test)231232> **These checks are about the shell, not the contents.** Every item below can233> pass on a document that argues nothing — that is not a defect in the list, it234> is its scope. The contents are gated by `spec-contract`, whose checker is the235> last item here and is the only one that reads what the doc says.236237- [ ] `<title>` and `<h1>` match (with doc-type prefix on `<title>`)238- [ ] `<p class="meta">` line present with author + date + status + risk239- [ ] Theme CSS is the `references/theme.css` content verbatim (or `<link>` to it if hosted) — no inline drift240- [ ] At least one `<div class="toc">` if the doc has ≥4 H2 sections241- [ ] All `tag` chips use one of the six canonical classes (info/ok/warn/bad/pick/cfd)242- [ ] Filename matches `YYYY-MM-DD-<slug>.html` (date = mtime, slug = topic)243- [ ] **Contents gated**: `python3 skills/governance/spec-contract/scripts/spec_check.py <doc>` exits 0, and the `spec-contract` rubric scores ≥11/15 with no axis at 0244245## References246247- Canonical theme source: `docs/specs/2026-05-22-broker-selection-cross-asset.html` lines 8–71 (extracted into `references/theme.css`)248- P18 audience rule: `CLAUDE.md` §P18 + `AGENTS.md` §P18249- trq212 thesis: *The Unreasonable Effectiveness of HTML* (May 2026, Claude Code team) — the empirical case behind P18250- Related skills: `bookkeeping` (Category-B render), `handoff` (companion MD artifact), `Webdesign` (product UI, disjoint)