feat: add Cap for Agents page and nav tab

[richiemcilroy]

What

Adds /agents, a marketing page for using Cap entirely from an AI agent, and an Agents tab in the site nav.

The page is built in the HomeTwo design language (Instrument Sans / Source Serif / DM Mono, cream card, grain, mode meshes, flat header + island) and reuses the existing agent scene from the homepage.

Sections

• Hero: "Give your agent a screen recorder." with a primary Copy the setup prompt button, a See how it works anchor, a copyable cap.so/agents pointer pill, the harness list, and the existing AGENT scene in the dark card.
• Set up Cap with one paste: the full setup prompt (same source as the docs, now in data/agent-prompt.ts) in a collapsible terminal card, three HowTo steps, and a "Point your agent here" card with a one-line prompt that tells an agent to read this page.
• Works in Claude Code, Codex, Cursor and OpenCode: platform-aware installer, per-harness tabs with the exact cap agents install commands, the OpenCode MCP entry, generic stdio config, what gets installed, and read-only verification commands. Every snippet has a copy button.
• Everything Cap does, from your agent: a new animated task reel (record, summarize, search, comment with confirmation, move, Loom import) driven by an rAF clock; clicking a capability row seeks the reel, the active row shows a progress line. Pauses off-screen, when the tab is hidden, and under prefers-reduced-motion.
• Prompts you can paste right now: six copyable example prompts from the docs.
• Safe to hand to an agent: the four safety boundaries from the docs plus a headless / CAP_API_KEY note.
• FAQ (9 questions) and a final CTA.

SEO

• Title Screen Recorder for AI Agents | Cap for Agents, description, keywords, canonical, OG/Twitter via buildMarketingMetadata (signed /api/og image verified).
• JSON-LD: BreadcrumbList, HowTo, FAQPage, SoftwareApplication.
• Internal links: desktop nav tab, mobile menu, footer, homepage Agents section button now links to /agents; llms.txt / llms-full.txt point agents at the page. /agents is picked up by the sitemap automatically.
• The page is written to be readable by agents: the full prompt is always in the DOM (the collapse is visual only) and the pointer card tells an agent that the prompt is its operating instruction set.

Other changes

• CAP_AGENT_PROMPT moved to apps/web/data/agent-prompt.ts and re-exported from the docs CopyablePrompt so both surfaces share one source of truth.
• HomeTwo/scenes/AgentScene.tsx: unbroken JSON output in the tool rows overflowed the viewport at phone widths (homepage too); overflow-wrap: anywhere on the output container fixes it.

Validation

• Biome clean on all touched files; anchor ids use the same biome-ignore pattern as Workflow.tsx / ModeWalkthrough.tsx.
• tsc --noEmit for apps/web: no errors in touched files (the 30 pre-existing errors are in untouched files such as workflows/edit-video.ts).
• __tests__/unit/agent-docs.test.ts: 14/14 pass.
• Playwright against a local dev server at 1440 / 1024 / 390: hydration with zero console errors, copy buttons write the clipboard, header island appears after the hero, harness tabs switch, capability rows seek the reel, the mobile menu lists Agents, no horizontal overflow at 390 / 360 / 320, and the no-JS render still contains the H1 and the full prompt. The nav fits at 1024 with the new tab.
• Homepage re-checked at 390 / 1440 after the shared scene change: no overflow, no errors.
• Compile-only production build + next start: /agents serves in production mode with all of the above passing at 1440 and 390 and zero console errors. A full next build currently dies on the pre-existing useContext prerender flake on /admin/reprocess-video and /messenger (neither touches these files), so static prerender of the route could not be confirmed locally.

Confidence Score: 5/5

The PR appears safe to merge; the previously reported contract, analytics, and lint-suppression issues are resolved.

Summary

• Shares the agent setup prompt between the documentation and marketing surfaces.
• Provides harness-specific installation instructions, safety guidance, examples, FAQ metadata, and copy controls.
• Adds an animated capability reel demonstrating recording, library operations, collaboration, and imports.
• Updates desktop, mobile, footer, homepage, and LLM-facing links to expose the page.
• The changes since the previous review correct the demonstrated CLI contracts and replace lint suppressions with shared anchor constants.

_{Reviews (3) · Last reviewed commit: "fix: show the real operation, context, a..."}

[superagent-security]

Superagent found 1 security concern(s).

[superagent-security]

P1: The agent setup prompt executes an unverified remote installer

The new agent prompt executes a network-fetched installer without integrity verification.

Use a versioned, signed installer and verify its checksum before execution.

AI prompt

Check if this security scanner issue is valid. If so, understand the root cause and fix it. If appropriate, update or add tests. Keep the change focused and preserve intended behavior.

<file name="apps/web/data/agent-prompt.ts">
<violation number="1" location="apps/web/data/agent-prompt.ts:21">
<priority>P1</priority>
<title>The agent setup prompt executes an unverified remote installer</title>
<evidence>The newly published agent-facing operating instructions tell an agent to fetch a shell installer over HTTPS and execute it directly when the CLI is missing. The downloaded script is not pinned, checksum-verified, or otherwise inspected before execution, so compromise or replacement of the endpoint would become arbitrary code execution on the user's machine.</evidence>
<recommendation>Replace the pipe-to-shell and PowerShell expression execution with a versioned installer artifact whose signature or SHA-256 is verified before execution. Prefer directing the user to an independently verified installation flow, and document the exact version and integrity value.</recommendation>
</violation>
</file>

[richiemcilroy]

hey @greptileai, please re-review the PR

[richiemcilroy]

hey @greptileai, please re-review the PR