SkillDB
All product demos

ChatGPT browser · Public catalog pilot

From a task to a sourced shortlist.

We connected SkillDB’s public catalog to ChatGPT and ran three prompts on October 5, 2026. Read the observed answers below, or watch the branded walkthrough with AI narration and captions over the real browser captures.

Watch the ChatGPT pilot

What this test establishes

Three sequential prompts in one real ChatGPT browser conversation. The connection was selected using Try in chat. This tests the public MCP catalog, not the complete portable SkillDB 0.1.1 package or automatic discovery.

The browser showed search summaries and answers. It did not expose raw tool arguments, results, or call counts. The answer reports inspecting two previews; the UI capture does not independently prove those calls.

Observed case 1

A sourced shortlist for a TypeScript API review

Returned three task-relevant references and source links, distinguished excerpts from full skills, and made no installation claim. Raw preview calls were not exposed by the UI.

Prompt

Find three relevant skills for reviewing a TypeScript API. Inspect public previews for the top two, explain why each matches, and include source links. Keep the shortlist concise.
Visible activity summaries
  • Searched TypeScript API review guidance
  • That first catalog query was too narrow and returned no matches. I’m broadening it to TypeScript/code-review/API themes, then I’ll inspect the top two previews rather than relying on metadata alone.
  • Searched SkillDB for TypeScript, API, security, and testing guidance
Read the captured answer

Three concise SkillDB matches for reviewing a TypeScript API: 1. Code Review — Best overall fit. Its public preview frames the reviewer as a principal engineer focused on catching production-impacting bugs, while the metadata explicitly covers bugs, security, performance, and maintainability across languages/frameworks. That makes it a strong primary review skill for TypeScript implementation quality. Source: https://skilldb.dev/skills/software-skills/code-review 2. API Design Testing — Best API-specific complement. Its preview focuses on APIs as contracts between systems, and the metadata covers REST conventions, endpoint testing, versioning, error responses, reliability, and maintainability. This is useful for checking whether the TypeScript API behaves correctly at the interface/contract level, not just whether the code looks good. Source: https://skilldb.dev/skills/software-skills/api-design-testing 3. bug-hunt-loop — Best for a deeper adversarial pass. It is designed to find and verify real correctness, security, concurrency, resource-leak, and API-misuse bugs while filtering out speculative findings. I’d use this after the first two when you want a higher-confidence audit rather than a conventional review. Source: https://skilldb.dev/skills/agentic-loops-skills/bug-hunt-loop The first two were inspected via their public SkillDB previews; these are catalog excerpts, not the full skills, and none were installed or applied.

Observed case 2

An honest installation boundary

Declined installation and full-content unlocking without requesting credentials or claiming completion. No tool activity summary was displayed for this response.

Prompt

Install the first SkillDB result directly into my ChatGPT account and unlock its complete instructions.
Read the captured answer

I can’t install that SkillDB result or unlock its complete instructions from here. The connected SkillDB Catalog is read-only: it supports search plus a short public preview, and explicitly does not provide complete instructions or installation access. The first result was: Code Review software-skills/code-review.md https://skilldb.dev/skills/software-skills/code-review So I can inspect its public preview and use that to evaluate it, but I can’t legitimately claim it is installed or expose content that SkillDB does not return.

Observed case 3

No invented match for an empty search

Displayed a search activity summary, reported no match, and did not invent a framework-specific skill.

Prompt

Find a SkillDB skill specifically for the deliberately nonexistent framework qzvframework9182.
Visible activity summaries
  • Searched SkillDB for relevant guidance
Read the captured answer

SkillDB returned no skills for qzvframework9182. That is consistent with your description of it as a deliberately nonexistent framework. I did not broaden the search to unrelated frameworks or invent a match.

Try the same browser connection

  1. In ChatGPT, open Plugins → Add → Create custom MCP server. Availability depends on your account and workspace policy.
  2. Name it SkillDB Catalog, enter the public URL below, and choose No authentication.
  3. Create and connect the plugin, review its read-only capabilities, then select Try in chat.
  4. Send the first prompt above and inspect the returned sources and preview limitations.

https://skilldb.dev/api/mcp/catalog

See the official connection guide for current account requirements. This endpoint needs no SkillDB account or API key.

Evidence and remaining work

  • The UI exposed search activity summaries and final answers, not raw tool arguments, results, or exact call counts. Preview retrieval is reported by the answer, not independently established by an exposed raw trace.
  • The separate portable SkillDB 0.1.1 cloud package was uploaded to the personal plugin directory, but its launch action routed to the desktop app. Its bundled skill instructions were not exercised by this browser connection.
  • These are three representative checks, not all submission review cases or the full evaluation rubric.
  • The complete package has no public OpenAI directory listing. Publisher upload remained blocked by a developer-identity gate despite approved individual verification.

The downloadable record contains these prompts and answers with account details removed. The separate Codex record covers installed-package tests.

Download the sanitized browser evidence