[ukimsanov]

What

• Three new pages, one per agent: Make videos with Claude Code, with Codex, with Cursor. Each one: install (plugin or skills), one message to paste, what the agent does, how to reopen the preview and render, the app or hosted-chat alternative, and what to check when something goes wrong.
• Where they live: Guides → Choose where to create → With your agent, next to the AI chat and design-tool pages.
• Who links to them: the quickstart's "With your agent" steps, Choose how to create, and Connect your agent in the Studio app pages.
• One fix on the quickstart: it said the agent "proposes a direction and asks for anything missing". Since the one-skill change (feat(skills)!: publish seven knowledge skills aligned with Studio and retire the workflow skills #5651) the skill builds a short video as asked and offers three storylines only for a longer film that tells a story (skills/hyperframes/SKILL.md), so the sentence now says that.

Why

People search by the agent they already use ("claude code video", "codex make video"), and each agent installs differently: Claude Code has a plugin, Codex reads ~/.agents/skills, Cursor reads project skills. Today all three share one generic block on the quickstart, so a Codex or Cursor user has to work out which parts apply to them.

How we'll know: Mintlify views on the three pages, and whether " + video" searches start landing on them in Search Console.

Related work

Third docs PR (Studio app pages #6223 and Start pages #6303 merged). Each page will get a recorded session of that agent making a video; that recording is listed in the tracker's "Media needs" tab and doesn't hold up the text.

How

• Fact check: 65 claims checked against main of hyperframes (CLI, skills/hyperframes/SKILL.md, plugin manifests, docs/guides/plugins.mdx, skills.mdx) and hyperframes-internal (packages/desktop/main/harnessCatalog.mjs for which agents the app runs: Claude Code and Codex yes, Cursor no). 15 were wrong in the 20 Oct drafts and are fixed: "Studio" → "browser editor", the outdated "picks a workflow" and "proposes a direction", and broken anchors (/help#troubleshooting, /guides/skills#… → /guides/plugins#…, connect-your-agent#codex). Checked outside the repo: the skills add picker lists Claude Code, Codex and Cursor (skills 1.7.4); Codex calls a skill with $name and reads ~/.agents/skills (OpenAI's Codex skills docs).
• No new components; plain Mintlify Markdown.

Test plan

• mint broken-links --check-redirects: no broken links; mint validate passes; bun run lint passes
• Manual testing performed: the three new pages plus quickstart, Choose how to create and Connect your agent loaded in Chrome on a local preview, 0 page errors; every internal link and anchor on them answers 200; the sidebar shows the "With your agent" group
• Documentation updated (if applicable)
• Unit tests added/updated (docs only)
• Comments follow CONTRIBUTING.md "Comments"

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
hyperframes	🟢 Ready	View Preview	Oct 25, 2026, 9:35 PM

💡 Tip: Enable Automations to automatically generate PRs for you.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

[jrusso1020]

Reviewed exact 4a25357. No blocking finding in this seven-file docs change.

Sampled factual checks against freshly fetched main 059a6e5 support the instructions:

• Claude's marketplace/plugin names and /hyperframes:hyperframes invocation match the manifests and plugin guide (docs/guides/with-claude-code.mdx:11–12,56). The current Claude plugin documentation also supports shell installation, reload and marketplace auto-update controls.
• Codex's shared store and $hyperframes invocation (with-codex.mdx:15–16,44) agree with the CLI installer and official OpenAI skill documentation. Cursor's project/global paths (with-cursor.mdx:38,93–95) agree with skills@1.7.4, the CLI mirror and Cursor's skill documentation.
• Independent project-scoped installs using cached skills@1.7.4, explicit Codex/Cursor targets and the exact local repository successfully place the selected hyperframes skill in each scratch project's .agents/skills. This exercises the selected installation path, not the interactive Core Skills picker, full remote install or global-user mutation.
• The CLI's core set/global fan-out, Node>=22 requirement, preview, render --output and doctor commands match current source. The current HFI harness catalog lists Claude Code and Codex as native app agents, with no Cursor harness.
• The storyline wording matches skills/hyperframes/SKILL.md:46: new story films over approximately 15s get three storylines; a 10s request is not promised that selection step. The skill retains its explicit file/no-stops exceptions.

Actual Mintlify local preview loads the three new guides plus Quickstart, Choose how to create and Connect your agent with no page errors. The Guides sidebar exposes With your agent and all three destinations. All six sampled navigation destinations/anchors resolve, including plugins#claude-code, plugins#cursor and quickstart#with-your-agent. mint validate and mint broken-links --check-redirects both pass.

Audited: all seven changed files, neighboring setup guides, sampled CLI/skill/plugin source contracts, current app harness catalog and rendered navigation/anchors. This is a sample of the 65 claimed facts, not an independent 65-claim certification. No signed-in provider session, paid generation, rendered-video acceptance, native app installation or global skills update performed; the author's lint claim was not independently rerun.

Need: yes, agent-specific setup paths · Size: focused docs addition · Reuse: existing plugin/skills reference guides and commands · Simpler: direct pages and ordinary links · Structure: fine; separate agent journeys justify the repeated short preview/render examples without a new component.

— hf-review

Verdict: COMMENT
Reasoning: Sampled install/skill behavior and the storyline correction agree with current source, and navigation/build controls pass. Approval requires James or a trusted stamper's go.

[miguel-heygen]

Reviewed exact 4a25357e9c6d83cb3f64b0bf88245e24e6e28990; no blocking finding in the seven-file documentation change.

The per-agent entry points preserve the distinct install/invocation contracts (docs/guides/with-claude-code.mdx:11, with-codex.mdx:15, with-cursor.mdx:25), and the three-storyline wording now matches the current skill's longer-story-film rule (skills/hyperframes/SKILL.md:46).

Audited: all seven changed files and authoritative git/API file lists; plugin manifests and plugin execution rules; CLI global-install/core-selection, Cursor mirror, Node prerequisite, preview/render/doctor contracts; neighboring installation guides. Independently checked official Claude, OpenAI and Cursor documentation. The locally available Desktop harness catalog at da853f91 supports Claude/Codex and excludes Cursor; this is a source check, not current packaged-app acceptance.

Executed evidence: all six affected public pages load on the published Mintlify PR preview; the sidebar reaches Cursor, Claude's standalone tab and Cursor's user-install tab work, and Cursor renders at 390px in dark theme without horizontal overflow. Forty-five local internal-link/anchor references resolve. No captured page exceptions. Two console/request errors are the existing site chrome's unauthenticated GitHub repository-metadata requests returning 403; the changed pages and tab contents render successfully. The downloaded current-head Validate docs log explicitly records successful mint validate and mint broken-links --check-redirects executions.

Trusting / limits: Rames's review5645429291 supplies the separately executed Codex/Cursor scratch-install controls; I did not repeat installs, certify all 65 author claims, start a signed-in agent session, generate/render a video, or install the native app. No product changes, dependencies, worktree or server created; task browser closed. Final exact-head snapshot: 24 checks succeeded, 37 skipped, none failed or pending. Skipped suites are not claimed as executed.

Verdict: APPROVE
Reasoning: The setup and storyline claims checked against source/vendor contracts agree, relevant documentation validation passed, and the published preview's changed navigation and installation variants work. No remaining blocking code or documentation finding.

— Magi