Voice, point of view, and editorial gate
The six FrankX stances, the order-of-truth prose standard, banned AI phrasing rules, and how check-frankx-voice.mjs runs in CI.
Starlight-Intelligence/starlight-design-intelligence/brand-packs/frankx/POINT_OF_VIEW.md, Starlight-Intelligence/starlight-design-intelligence/brand-packs/frankx/COPY.md, Starlight-Intelligence/starlight-design-intelligence/scripts/check-frankx-voice.mjs, frankxai/frankx.ai-vercel-website/scripts/audit-ai-slop.mjsFrankX copy follows a single posture defined in starlight-design-intelligence/brand-packs/frankx/POINT_OF_VIEW.md: hope with receipts. Public writing speaks founder-to-founder with direct technical mechanism, observable proof, and explicit boundaries.
The six FrankX stances (POINT_OF_VIEW.md)
Every flagship essay, product page, or AI Architect newsletter issue anchors its argument in one of six verified stances:
| Stance | Core claim | Required proof | Boundary ("Never say") |
|---|---|---|---|
| 1. One person can now build what a team built | A single operator with well-run agents can ship products, research, and revenue loops that previously required several people. | A named artifact with its build time, cost, and the checks it passed. | Weekend millionaire promises or any income/earnings claim. |
| 2. From vibe coding to agentic engineering | Prompting a demo is a good start; durable products require version control, tests, evals, cost caps, permissions, and a human stop switch. | A before-and-after on a real build showing what broke and which check caught it. | Dismissing vibe coders; show them the next engineering rung instead. |
| 3. Explore with curiosity, build your own defenses | The capability that gives a founder leverage also gives attackers leverage; operators and families need account security, data ownership, and verification habits. | A concrete defense a reader can apply today, the threat it answers, and its limits. | Doom framing, job-loss fear hooks, or untested security advice. |
| 4. The frontier is a shared project | No single lab or company gets AI right alone; study primary lab releases, credit them, test them, and report what works. | Linked and dated primary sources (papers, system cards, release notes). | Tribal model wars ("X just killed Y") or unattributed forecasts. |
| 5. Ambition is welcome when earned by value | Building revenue and wealth with AI tools is a valid goal when the money follows technology and design buyers would miss. | What the product does for a named buyer and its honest price once live. | Passive-income promises, fake scarcity, or earnings screenshots as proof. |
| 6. AI will help us find answers we could not find alone | The largest long-term return comes from science, medicine, energy, climate, and education. | A cited discovery, benchmark, or workflow with clear separation between measured results and open problems. | Medical advice, miracle cures, or treating speculative science as settled. |
Rules for discussing AGI and frontier forecasts
When a piece covers AGI or superintelligence, POINT_OF_VIEW.md enforces eight calibration rules:
- Attribute every forecast with the forecaster's name, date, and venue.
- Pair every capability claim with its physical, data, regulatory, or cost bottleneck.
- Prefer scenario ranges (plateau, continue, accelerate) over single dates.
- State both error costs (acting too early vs. waiting too long) before advising.
- Separate primary source summaries from interpretation on the page.
- Disclose commercial interest beside the claim, including FrankX products.
- Argue from observable effects rather than unprovable claims about consciousness.
- Document changed views by quoting the earlier position and the new evidence.
Order of truth and editorial tests (COPY.md)
On any high-value page, the first 120 words must establish three things in order:
- Why this page exists for this reader.
- What Frank actually saw, made, tested, or learned.
- The smallest honest proposition for a next move.
Before copy ships, it must pass four editorial tests defined in brand-packs/frankx/COPY.md:
- Name-swap test: Replace the person or product name. If the paragraph still works unchanged, it is generic and cannot ship.
- Builder-taxonomy test: Remove words like
agent,skill,infrastructure,ecosystem,sovereign,substrate, andauthority graph. If the meaning disappears, the sentence was hiding behind taxonomy. - Concrete-anchor test: Every important claim needs one observable anchor (an artifact, commit, decision, metric, quotation, or working behavior).
- Deletion test: Delete the grandest adjective and most abstract noun. If the line improves, keep the deletion.
All public headings and interface labels use sentence case. Frank's title is always AI Architect (never expanded variants), and Arcanea lore terms (Luminor, Guardians, Gates, Shinkami) are quarantined from FrankX surfaces.
Mechanical voice & anti-slop gates in CI
Two deterministic Node scripts enforce these rules automatically in local pre-push checks and GitHub Actions:
1. check-frankx-voice.mjs (starlight-design-intelligence)
Run node scripts/check-frankx-voice.mjs <file> (or pass --template when linting specimen templates with bracket slots). The linter masks fenced code blocks and checks prose line-by-line against 11 line rules and 2 document-level rules:
| Rule ID | Severity | What it catches |
|---|---|---|
em-dash | error | Any em dash (U+2014) in FrankX prose. |
unattributed-forecast | error | Frontier/AGI future claims missing an attribution verb, named source, and statement date. |
income-claim | error | Passive-income phrases, multi-figure income promises, or "quit your day job" hooks. |
inevitability | error | Fear or inevitability framing ("no stopping this", "left behind", "adapt or die"). |
unpopular-opinion | error | Lines opening with Unpopular opinion:. |
arcanea-diction | error | Arcanea proper nouns (Arcanea, Arcanean, Luminor, Luminors) leaking into FrankX copy. |
unfilled-slot | error | Unfilled [bracket slots] in non-template copy. |
numbered-opener | error | Document opening with 7 lessons/tips/ways... or 1/n thread headers. |
golden-age-repeat | error | Using the phrase "golden age" more than once in a single piece. |
arcanea-name-check | warn | Ambiguous canon words (Lumina, Nero) for human confirmation. |
hype-verb | warn | Hype verbs (unlock, reimagine, elevate, 10x your) that lack a measurable object. |
contrast-formula | warn | Rhetorical negative parallelisms (not just X, but Y or X, not Y). |
stock-phrase | warn | Stock model tells (delve, tapestry, in today's fast-paced, ever-evolving landscape). |
2. audit-ai-slop.mjs (frankx.ai-vercel-website)
Inside frankx.ai-vercel-website, pnpm run ai-slop:audit:strict runs node scripts/audit-ai-slop.mjs --strict across app/, components/, content/, and lib/ (paired with node scripts/audit-editorial-signals.mjs for weighted prose signal scoring):
# Fail-closed CI check across app/, components/, content/, and lib/
pnpm run ai-slop:audit:strict
# Inspect weighted editorial signals on a specific blog post
node scripts/audit-editorial-signals.mjs --strict --file content/blog/example.mdxSkills library & MCP doctor
How the 124 installable skills in claude-skills-library and the @frankxai/mcp-doctor CLI are structured, resolved, and installed.
Digital products, blueprints, and research
Structure of the FrankX product catalog in data/products.json, the research publication gate, and the frankx.ai/docs proxy route.