content-gate: a pre-publish gate for AI-written posts
content-gate turns Google's people-first guidance into a pre-publish gate: project guidelines loaded first, five spam hard-stops, a pass or fail verdict.
Skill details
npx skills add vikingmute/content-gate --skill content-gate
AI multiplies how much content a team ships, and volume slides toward search-engine-first filler: near-duplicate pages, date bumps, overpromised titles — exactly what Google’s spam policies call out. content-gate (vikingmute/content-gate, by Viking Zhang) turns the official guidance into a gate: before drafting, rewriting, reviewing or approving any blog or guide article meant to rank, the agent runs a people-first check, and what fails doesn’t ship. It is not a keyword tool; the README says it plainly — quality over page count.
The order it works in
I read the full SKILL.md; the sequence sets rules before scores. Step one finds the project’s own editorial doc (CONTENT_GUIDELINES.md or similar) and treats it as the hard gate — the skill only supplies the shared procedure. Then one question decides everything: who is helped, and what task can a reader finish? If the honest answer is “rank for search”, the article does not publish. The Who / How / Why lenses follow (clear attribution, steps that match the real UI, AI assistance disclosed where a reader would ask), and the run ends in a verdict — pass, conditional with a named fix plan, or fail — written into the PR or review notes.
What it constrains
- Five spam hard-stops, any one fails the piece: writing mainly to manipulate rankings, batch-producing near-identical thin pages, scraped or synonymized low-value translation, keyword stuffing or doorway pages, and date changes with no substantive update.
- People first, search second: a direct visitor must find it useful, a reader must finish a concrete task, first-hand traces (real steps, data, screenshots) earn their place, and an accurate title is the floor.
- AI drafts need deep edits: agent-written copy must be edited past the hollow-shell stage, disclosed where a reader would reasonably ask, and never bylined as “AI”.
- No invented numbers: no traffic forecasts, keyword volumes or competitor scores to argue a piece into passing.
Who it’s for
Teams that mass-produce blogs and docs with AI get the most out of it — install it into Claude Code or Codex and require the gate report on every content PR:
npx skills add vikingmute/content-gate
The honest catches: the repo carries no LICENSE file (checked September 29), so formally all rights are reserved — worth confirming with the author before wiring it into a team process; the skill is pure instruction text with no executable checker, so enforcement depends on the model actually following it; and it governs whether to publish, not keyword research or link building. Paired with AiToEarn it bookends the pipeline: that one distributes widely, this one keeps you from shipping the wrong thing.