Skip to main content
SkillDB LabsProduct walkthrough

From finding a skill
to putting it to work.

A closer look at the new tools: focused starter bundles, a workbench for your drafts, and libraries that keep useful references within reach.

These workflows are in development. Explore what each one does today and where a real-world evaluation still comes next.

Guided walkthrough · Actual product screens1:17

SkillDB Labs: a guided product walkthrough

Explore starter bundles, the draft workbench, account-library entry points, workflow packages, and the timeline.

Edited from actual public SkillDB screens, with a captured timeline interaction. Still screens are held for reading and idle gaps are shortened. The workbench uses its labeled sample. Account features are shown signed out. This is not a ChatGPT recording or a directory-publication claim.

Read the full transcript

The same reviewed text is available through the player’s English caption track. Download captions (.vtt)

  1. 0:00

    Explore SkillDB Labs: new tools for choosing skills and keeping evidence.

  2. 0:07

    Choose a React app, API service, accessibility review, or data workflow.

  3. 0:15

    Review each role, overlap, and limitation. Exports contain references; they do not install skills.

  4. 0:24

    This sample has intentional conflicts. Text checks show the exact lines behind each suggestion.

  5. 0:33

    Select a line reference to focus the instruction that needs review.

  6. 0:38

    Keep normal, edge, and adversarial cases. Record real outcomes; the workbench does not run a model.

  7. 0:45

    Account collections save public references and support comparison. This is the signed-out entry screen.

  8. 0:51

    Team libraries use account membership and roles. Sign-in is required; no private workspace is shown.

  9. 0:57

    Code review and form accessibility packages show their scope. Directory publication remains pending.

  10. 1:04

    Daily and monthly controls update the real chart. This interaction is captured from the live site.

  11. 1:12

    Try the tools at skilldb.dev/labs. Development previews make their current limits visible.

Codex test transcript · Actual captured results0:45

SkillDB in Codex: the recorded test transcript

Follow a real installed-plugin pilot: live discovery, preview limits, unsupported installation, and an empty search.

A guided view of the sanitized report from three real Codex CLI 0.160.0 runs with SkillDB plugin 0.1.0. The responses and tool summaries are rendered for reading on the evidence page. This is not footage of ChatGPT or the Codex interface, and does not demonstrate directory publication. The separate ChatGPT catalog pilot appears below.

Read the full transcript

The same reviewed text is available through the player’s English caption track. Download captions (.vtt)

  1. 0:00

    Recorded Codex CLI events, rendered for reading. This is a test transcript, not a ChatGPT screen recording.

  2. 0:08

    The installed instruction file was read. Codex called the live SkillDB catalog search and preview tools.

  3. 0:16

    The response separates catalog descriptions from limited previews. No complete skill was installed or applied.

  4. 0:28

    The installation request was declined. No account or installation tool was called.

  5. 0:37

    The empty-result case returned no invented skills. Read the prompts and evidence at skilldb.dev/demo/codex-pilot.

ChatGPT catalog pilot · AI-narrated walkthrough1:05

SkillDB in ChatGPT: find your next useful skill

A narrated walkthrough of task-based discovery, sourced previews, and clear capability limits in a real ChatGPT browser pilot.

A branded walkthrough made from actual screenshots of the custom SkillDB Catalog MCP connection in ChatGPT. Screens are cropped for readability and held with edited pauses; this is not a continuous recording. AI-generated stock-voice narration describes the observed UI and was added in editing. Browser MCP connection; full package launch is separate. Visible search summaries and answers do not establish raw invocation counts. No skill installation, private access, or directory approval is demonstrated.

Read the full transcript

The same reviewed text is available through the player’s English caption track. Download captions (.vtt)

  1. 0:00

    Your next project doesn't need to start from scratch. SkillDB helps you find reusable skills for AI agents.

  2. 0:07

    Bring that discovery into ChatGPT with SkillDB's public catalog connection.

  3. 0:13

    Start with a task: find three skills for reviewing a TypeScript API, with public previews and source links.

  4. 0:21

    The visible search summary shows ChatGPT broadening a narrow query to find relevant guidance.

  5. 0:27

    The answer recommends Code Review, API Design Testing, and bug-hunt-loop. It explains their fit and links the sources. These previews are excerpts, not complete instructions.

  6. 0:38

    Ask to install a skill or unlock its full content, and ChatGPT explains the limit: this connection supports read-only discovery.

  7. 0:46

    For a deliberately nonexistent framework, the answer reports no match. It says so without inventing a skill.

  8. 0:54

    Find your next useful skill at skilldb.dev. Explore the catalog, read the pilot evidence, and choose what to try next.

Explore at your own pace

Six parts of a useful workflow.

Each step links to the real product. Account features require sign-in; the draft workbench runs locally in your browser.

01Public discovery · Host pilot

Find a skill for the task

Search the public catalog, inspect a preview, and follow the source. Discovery helps you choose what to read; it does not apply a skill automatically.

The discovery pilot exposes public metadata and previews.

Explore agent discovery
02Development preview

Start with a focused bundle

Review a small set of skills for a React app, API service, accessibility review, or data workflow. Each reference has a stated role.

Downloads are reference plans. Installation is a separate step.

Browse starter bundles
03Development preview · Sign-in required

Keep team practices together

Organize workspace-owned practices with explicit members and roles, and keep private material behind account and workspace checks.

Authenticated API checks passed. Multi-account browser and paid Studio validation remain pending.

Explore team libraries
04Local draft tools · In development

Improve a draft with evidence

Inspect explained text checks, define normal and edge cases, and record manual outcomes with the exact draft and test inputs.

No model is run here. Drafts stay local unless you save or export them.

Open the workbench
05Account collections · In development

Compare a useful shortlist

Save public references to your account, compare their stated coverage, and copy the comparison into your conversation.

Descriptions and length provide context, not quality ratings.

Open collections
06Package preparation · Distribution pending

Inspect focused workflow plugins

Review focused code-review and accessibility workflows, their source references, expected outputs, and evaluation plans.

Curated workflow drafts are separate from the public discovery plugin.

Browse workflow drafts

OpenAI plugin pilot

See what was actually tested.

The public discovery plugin searches catalog metadata and previews. It is separate from the curated workflow packages and does not access private team content.

ChatGPT catalog pilot

A branded walkthrough made from actual screenshots of the custom SkillDB Catalog MCP connection in ChatGPT. Screens are cropped for readability and held with edited pauses; this is not a continuous recording. AI-generated stock-voice narration describes the observed UI and was added in editing. Browser MCP connection; full package launch is separate. Visible search summaries and answers do not establish raw invocation counts. No skill installation, private access, or directory approval is demonstrated.

View the captured evidenceRead the verified Codex test transcriptRead the ChatGPT browser pilot and try it yourself

Development demonstrations are not a public directory listing or a guarantee of results on every task.

Bring a task of your own.

Inspect the source, choose a workflow, and keep the evidence for what helps.