Editorial Policy — How We Test and Publish — Skills and Servers
Subscribe

Editorial Policy

How we produce every page — testing standards, refresh cadence, correction handling, byline attribution, and the strict boundary between editorial and advertising.

SS Skills and Servers Editorial Last updated August 15, 2026

Our editorial principles

Skills and Servers exists to help developers find the right AI primitives for the job — fast, without vendor spin, without half-tested links. Every page we publish is held to five standards:

  1. Tested. Every Skill has been loaded into at least one provider and run against real inputs. Every MCP server has been installed and its main tools called. If we have not tested something, we do not publish it.
  2. Provider-agnostic. We cover the whole ecosystem — Anthropic, OpenAI, Google, Meta, Mistral, Cohere, xAI, DeepSeek — and we mark clearly where support varies.
  3. Attributed. Every page carries a byline (Ahmed R. or Sana K.) and a "Last tested" date. You can see who verified what and when.
  4. Sourced. Claims about vendor behaviour link to the vendor's own docs. Claims about performance link to our test methodology or a published benchmark. We do not make things up.
  5. Independent. No AI provider pays us for coverage. Sponsored content is always labelled per our Advertising Disclosure.

How we test a Skill or MCP server

For a Skill:

  1. We load the Skill into Anthropic Claude Desktop or the equivalent client. Where the Skill is meant to run in Claude Code, Cursor, Cline, or another agentic client, we test it there specifically.
  2. We run the Skill against at least three representative inputs — an easy case, a typical case, and an edge case.
  3. We cross-test on OpenAI's Skills implementation and Google Gemini's implementation where the Skill is portable. Any behaviour differences get called out on the page.
  4. Any code sample on the page is run — not just written. If it does not run, it does not ship.

For an MCP server:

  1. We install the server from its official source using the documented instructions.
  2. We configure it against a real backing service (a real Postgres database, a real Slack workspace, a real GitHub repo, etc.) using our test accounts.
  3. We call every documented tool the server exposes, verify the response shape, and note any surprises.
  4. We test the recommended auth flow.
  5. We attempt to break it — malformed input, oversized requests, expired credentials — to document failure modes.

Corrections and updates

The AI ecosystem changes fast. When a Skill breaks, an MCP server changes its schema, or a vendor pushes a breaking change, we want to know as soon as you do.

Two ways to flag a problem:

We aim to acknowledge corrections within 1 business day and to update the affected page within 3 business days. When we make a material correction we note the update on the page with the date and a brief description of what changed.

Refresh cadence

Every published page is on a refresh schedule:

  • Every 90 days: we re-verify that documented URLs, code samples, and configuration snippets still work.
  • On vendor releases: when Anthropic, OpenAI, or Google ships a new model or a change to their Skills / MCP implementation, we sweep affected pages within 14 days.
  • On community reports: when a reader flags a broken entry, we fix it as described above.

The "Last tested" date at the top of every page reflects the most recent full re-verification.

Attribution and bylines

Every page carries the byline of the researcher who owned the test and the write-up. Our two lead researchers, Ahmed R. and Sana K., rotate ownership across the directory. Where an outside contributor's PR is merged, the contributor is credited alongside the editorial byline.

Sponsored or affiliate-related content is always labelled and does not carry a personal byline — instead it is attributed to "Skills and Servers Editorial".

Where we use AI in producing content

We use AI models — Claude, OpenAI, Gemini — as writing and research aids in the same way a technical writer uses Grammarly or Google. We do not publish AI-generated text without a human editor rewriting, verifying, and taking responsibility for it. Specifically:

  • No claim reaches a published page without a human editor confirming it against a primary source.
  • No code snippet reaches a published page without a human running it.
  • No comparison ("Skill A beats Skill B at X") reaches a published page without a human performing the comparison.

Our bylined researchers are humans, not personas. Ahmed R. and Sana K. are real members of our team.

The advertising / editorial boundary

Advertising, sponsorships, and affiliate revenue keep the site free to read. But they do not influence editorial:

  • An AI provider running display ads on our pages does not get better coverage.
  • An MCP server vendor running a newsletter sponsor slot does not get moved up in category rankings.
  • A sponsored placement is always labelled and always sits outside the editorial ranking of unsponsored entries.

See our Advertising Disclosure for the specifics on how we handle every commercial relationship.

Contact editorial

Feedback, tips, corrections, or story ideas: editorial@skillsandservers.com. We read everything and reply to most emails within 1 business day.

Related pages

Share with

Twitter/XPinterestRedditMastodonThreadsEmailPrintFlipboardFlipboardHacker NewsLineWeChatSnapchatMixMixPocketGoogle MessagesSMSTumblrTumblrXingXINGVKVKontakteBufferBufferYummlyNextdoorGrocyGabFacebookWhatsAppLinkedInTelegramMessengerBluesky