# Movemental Publish Gate > Pre-publication quality gate for research and authoritative content headed to movemental.ai. Run before posting any article, research synthesis, field guide, footnote source, /voices or /footnotes claim, or long-form doc. Checks ten red-light gates — AI-slop and register-jargon, citation and evidence integrity, factual/hallucination risk, doctrine drift (movement leaders vs orgs, "Trusted voices"), EEAT self-evidence, mechanical correctness, and legal/sensitivity — and returns a RED / YELLOW / GREEN verdict with the exact blocking items and which fix skill to run. A slop, mistake, and quality-control checker that decides "is this safe to publish yet?" Use whenever the user says publish / post / ship / "is this ready" / "check this before it goes live" / "QC this" about Movemental content. Routes to plain-prose, movemental-prose, and movemental-narrative-audit rather than duplicating them. - Skill: `joshuashepherd/movemental-publish-gate` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add joshuashepherd/movemental-publish-gate` - Raw SKILL.md: https://api.skillmd.com/api/skills/joshuashepherd/movemental-publish-gate/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Security - Author: JoshuaShepherd (https://skillmd.com/u/joshuashepherd) - Updated: 2026-09-08 - Page: https://skillmd.com/skills/joshuashepherd/movemental-publish-gate --- Run the publication gate on Movemental content: $ARGUMENTS `$ARGUMENTS` is a file path, a directory, or pasted prose. Prefix with `gate:` for verdict only (default), or `fix:` to run the routed fix skills after reporting. If empty, ask what to gate. This skill answers one question: **is this safe to publish on movemental.ai yet, and if not, exactly what blocks it?** It does not rewrite prose itself — it diagnoses, decides, and routes. The specialist skills do the fixing. --- ## How to run it 1. **Read the target** in full, including frontmatter. Note `audience_tier` / `canon_section` / `slug` / `author` / where it will live (`docs/articles/`, `docs/build/research/articles/`, `/footnotes`, `/voices`, home-tier). 2. **Walk all ten gates below in order.** Each gate returns one of: **RED** (blocks publication), **YELLOW** (fix strongly recommended, author may override with reason), **GREEN** (clear). Do not skip a gate because the piece "looks clean" — the slop and citation gates fail most often on polished-looking drafts. 3. **Apply the hard gates** (bottom of file). Any one hard-gate hit forces an overall **RED** regardless of the other nine. 4. **Emit the verdict block** (template at the bottom). Be specific: quote the offending line, name the file:line, and name the exact next action. 5. If invoked with `fix:`, after reporting, launch the routed skills for each RED/YELLOW gate (see the routing column). Higher public surfaces raise the bar. A `/footnotes` claim or a home-tier stat is held to the strictest reading of every gate; an internal `raw/` research note is held mainly to citation integrity and doctrine. --- ## The ten gates ### 1. Register & AI-shaped prose → does `plain-prose` / `movemental-prose` need to run? The user's first question is always "does the prose skill need to run?" Answer it here. Scan for the **register markers** (see `references/red-light-checklist.md` §1, and the watchlist that ships with `plain-prose`): substrate, scaffold, leverage, surface (v.), artifact, load-bearing, robust, orchestrate, operationalize, unlock, north star, flywheel, unpack, interrogate, problematize, deeply/fundamentally, "it's worth noting," "at the end of the day." Run the **costume-or-term-of-art test** before flagging — a real term of art in its home field is fine. Also scan for **AI cadence tells**: stacked tricolons, symmetrical "It's not X, it's Y" repeated, em-dash overload, hedge stacking, a conclusion that only restates the intro, "In today's rapidly evolving…", title-cased section headers, bullet lists where prose belongs. - **RED** — register markers are load-bearing (the argument leans on costume words) OR three or more distinct AI-cadence tells. → Route to **`plain-prose`** (register/jargon) then **`movemental-prose`** (Movemental line-level cadence). - **YELLOW** — a handful of swappable costume words or one or two cadence tells. → Recommend `plain-prose`. - **GREEN** — reads like a person talking to one senior leader. ### 2. Citation & evidence integrity (EEAT dimension B) The backbone. Every **material or empirical** claim that a reader could act on or quote must be traceable. Label each load-bearing claim: **Sourced** (author/date/publication/URL) · **Attributed** (named org, no specific report) · **Unsourced** · **Contradicted** (better evidence weakens it). - **RED** — any load-bearing claim is **Unsourced** on a public surface; a **Contradicted** or fabricated-precision stat is used (e.g. a bare "68%", "studies show," broken-telephone figures); empirical claims presented as theological/ethical certainty without a layer label. - **YELLOW** — claims are Attributed but not Sourced; gaps not yet marked "to verify." - **GREEN** — material facts Sourced, limits stated, primary + synthesis clearly separated. - Reference: `docs/build/notes/eeat-research-content-qualification-rubric.md` (movemental-ai repo) Dimension B and the per-claim confidence table; `docs/research/authoritative-sources-ai-nonprofits-faith-formation.md`. ### 3. Factual accuracy & hallucination Names, dates, book/article titles, org names, quotes, and links must be real and verifiable. AI drafts invent plausible titles, misattribute quotes, and fabricate URLs. - **RED** — a quote/attribution/figure can't be traced; an invented or misremembered book title, org, or person; a dead or fabricated link; a real person assigned words or positions they didn't say. - **YELLOW** — a citation that's probably right but unverified; a paraphrase that risks misquoting. - Verify links with `WebFetch` or `Bash` (curl -I) when in doubt. Cross-check named entities against the leader research and corpus before trusting them. ### 4. Doctrine alignment → does `movemental-narrative-audit` need to run? Check against Movemental canon (see `references/red-light-checklist.md` §4 and the canonical doctrine doc). - **RED (hard gate)** — **doctrine drift**: movement leaders placed as a fourth audience funnel/sibling card; any recruiting/roster/"nominate a leader" framing; the public label for the movement-leader surface is anything other than **"Trusted voices"** (not "Committed voices," "Scenius," "ambassadors," "partners"); "Scenius" used as a public H1; the two-intelligences / six-stage / infrastructure model misstated. - **YELLOW** — framework vocabulary drifts but the argument is recoverable. → Route to **`movemental-narrative-audit`**. - Canon: `docs/build/strategy/movement-leaders-as-ecosystem-layer.md`; CLAUDE.md "Canonical doctrine." ### 5. EEAT self-evidence Does the piece earn Experience, Expertise, Authoritativeness, Trust — or just assert them? Honest limits and counter-arguments stated? Author identifiable? Quotable, stable claims a machine could lift? - **RED** — makes authority claims it doesn't back; no honest limits on a piece whose whole job is credibility; anonymous on a surface that needs a named voice. - **YELLOW** — thin counter-argument pass; few extractable claims. Reference EEAT rubric Dimensions C and E. ### 6. Mechanical correctness - **RED** — leftover placeholders (`[TODO]`, `[RE-VALIDATE]`, `lorem`, `TK`, `XXX`, ``); broken internal links; missing/invalid frontmatter where the surface contract requires it (`src/lib/articles-schema.ts`); a duplicate of existing canon (grep the articles/research index before publishing a "new" piece). - **YELLOW** — heading hierarchy skips, prose width, spelling/typos, markdown that renders wrong. ### 7. Legal / sensitivity / pre-decision - **RED (hard gate)** — persona/seat entities presented as real leaders; unconfirmed pricing, scholarships, partnerships, or outcomes stated as fact; anything that "cannot be quoted without misleading" (EEAT Dimension E ≤ 4); private/internal strategy leaking onto a public surface without redaction. ### 8. Scope & audience fit - **YELLOW** — leader dossier content pushed into the *org* research series (different product); marketing copy dressed as research; internal ops notes promoted to a public article. Cap and redirect rather than publish in the wrong place. ### 9. Originality & duplication - **RED** — re-argues "settled thinking" (corpus-audit Part 3) without new primary evidence; near-duplicate of an existing article. → Mine for footnotes instead of publishing. ### 10. Surface readiness (the wiring) - **YELLOW** — claims not yet mapped to the `claim` + `cite` + `footnote` registry shape (`src/lib/citations/eeat-registry.ts`); no stable slug; internal links to canon not named. Not a publication blocker by itself, but list it so it isn't forgotten. --- ## Hard gates — any one forces overall RED 1. A **load-bearing claim is Unsourced**, or a **Contradicted / fabricated stat** appears, on a public surface (Gate 2). 2. An **unverifiable quote, attribution, title, or link** (Gate 3). 3. **Doctrine drift** — fourth funnel, recruiting/roster, wrong public label, "Scenius" as public H1 (Gate 4). 4. **Placeholders / RE-VALIDATE markers** left in the body (Gate 6). 5. **Persona/unconfirmed entities or pre-decision facts** stated as real (Gate 7). A piece can be beautifully written and still RED on any of these. Prose polish never overrides an integrity gate. --- ## Verdict block (emit this) ```text PUBLISH GATE — [path or title] Surface: [docs/articles | /footnotes | /voices | research raw | home-tier | …] Overall: 🔴 RED — do not publish | 🟡 YELLOW — publish with fixes | 🟢 GREEN — clear Gate Light Blocking finding (file:line) Route / fix 1 Register & AI prose 🔴/🟡/🟢 … plain-prose · movemental-prose 2 Citation integrity … … (fix sourcing) · EEAT rubric B 3 Factual / hallucination … … verify before publish 4 Doctrine alignment … … movemental-narrative-audit 5 EEAT self-evidence … … 6 Mechanical correctness … … 7 Legal / sensitivity … … 8 Scope & audience fit … … 9 Originality / duplication … … 10 Surface readiness … … BLOCKERS (must clear before publish): - … NEXT ACTION: [the single most important thing, named precisely — e.g. "Run plain-prose on ¶2–5, then source the '68%' stat in Gate 2, then re-gate."] ``` When unsure between two lights, take the stricter one and say why. The gate exists to keep slop and mistakes off movemental.ai — false-green is the only truly costly error. --- ## What this skill is not - Not a rewriter. It diagnoses and routes; **`plain-prose`** and **`movemental-prose`** do prose fixes, **`movemental-narrative-audit`** fixes doctrine/argument. - Not the EEAT *scoring* rubric — that 0–100 model (`eeat-research-content-qualification-rubric.md`) decides *whether a corpus doc is worth promoting*. This gate decides *whether a piece already chosen for publication is safe to ship*. Use the rubric to pick; use this gate to clear. - Not `alan-voice` / `article-audit` — if the piece is Alan-Hirsch-corpus voice, stop and route there instead.