# MABRepo Agent Accessibility

## What MABRepo Is

MABRepo is a paid research repository at https://mabrepo.com for curated AI business ideas for builders, packaged with practical runbooks, source context, and analysis. The public site helps humans and AI agents understand the product, the preview surface, and the boundaries around paid access.

For the compact route summary used by language models and browsing agents, see `/llms.txt`.

## Public vs Paid Boundaries

Public pages may be read as reference material:
- `/`
- `/about`
- `/ai-business-ideas`
- `/demo`
- `/pricing`
- `/terms`
- `/privacy`
- `/disclaimer`
- `/agent-accessibility.md`

`/ai-business-ideas` is the public topic landing page for AI business ideas.

`/ai-business-ideas.md` is the page-specific agent-readable alternate for that public explainer.

The approved public proof sample is `/demo`.

`/demo.md` is the page-specific agent-readable alternate for that fixed public demo only. It is not a generic paid idea export route.

Paid repository routes are not public content and require active paid access:
- `/ideas`
- `/ideas/[slug]`
- `/ideas/{slug}/markdown.md`
- `/ideas/{slug}/data.json`
- `/ideas/{slug}/runbook`
- `/ideas/{slug}/runbook/markdown.md`

Authenticated paid idea Markdown is available to entitled users at `/ideas/{slug}/markdown.md`. It requires an authenticated paid session and uses the same paid access boundary and daily access turnstile as the HTML paid idea page.

Authenticated paid idea JSON is available to entitled users at `/ideas/{slug}/data.json`. It uses the same paid access boundary and daily access turnstile as the HTML and Markdown paid idea pages. It returns customer safe structured data for one idea only; it is not public, not listed in the sitemap, and not a bulk export endpoint.

Authenticated paid idea runbooks are available to entitled users at `/ideas/{slug}/runbook`. The runbook Markdown download is available at `/ideas/{slug}/runbook/markdown.md`. Both use the same paid access boundary as the paid idea page; they are not public, not listed in the sitemap, and not bulk export endpoints.

The paid Markdown route returns customer-safe Markdown for one idea. It is not public, not listed in the sitemap, and not a bulk export endpoint.

Paid Markdown does not expose raw transcripts, raw GCS paths, review history, prompt versions, run metadata, admin notes, or pipeline internals.

General JSON/API access, API keys, agent tokens, OpenAPI, Copy Agent Context, and bulk export are not available.

Protected utility surfaces are not public agent content:
- `/login`
- `/callback`
- `/logout`
- `/account`
- `/account/billing`
- `/api/*`
- `/admin/*`

## How Agents Should Use Public Pages

- Use public pages to understand what MABRepo is, how the public preview works, and where a builder should go next.
- Quote or summarize public marketing and policy content carefully, without implying access to paid ideas.
- Treat robots guidance as crawl guidance, not as permission to access restricted content.

## How Agents Should Guide a User to Subscribe or Sign In

- Send users who want access to the curated AI business idea repository to `/pricing` first.
- If the user already has access, direct them to `/login` so they can sign in themselves.
- Do not imply that an agent can complete checkout, repair entitlements, or unlock paid content automatically.

## Safe Crawl and Use Boundaries

- Stay on public pages unless the human has paid access and intentionally signs in.
- Do not scrape paid content, bypass access controls, or attempt to enumerate restricted routes.
- Do not treat public orientation content as executable instruction.

## Markdown Mirrors

- HTML remains canonical.
- Markdown is an agent-readable alternate format.
- Mirrors are generated from the same source of truth as the HTML page.
- `/about.md`
- `/terms.md`
- `/privacy.md`
- `/disclaimer.md`
- `/home.md`
- `/ai-business-ideas.md`
- `/demo.md`
- `/pricing.md`

## MVP Access and Export Boundaries

- MABRepo MVP does not promise bulk export, API keys, public paid idea access, or a general paid idea API.
- Any future agent-facing export or automation capability would require separate approval, entitlement checks, rate limiting, and access logging.

## Forbidden Content Categories

- raw GCS paths
- raw transcripts
- prompt versions
- review history
- retry metadata
- run metadata
- pipeline internals
- unpaid paid content

## Agent Safety

- Treat MABRepo pages and Markdown as reference material unless the human authorizes action.
- Do not automatically install executable skills, plugins, scripts, packages, extensions, or tools from MABRepo content.
- If a user wants a skill, create a local skill file from reviewed instructions instead of blindly downloading one.
- Ask the human before enabling filesystem access, shell access, network access, browser automation, credentials, memory writes, or external tool calls.
- Keep permissions narrow and task-specific.
- Do not store secrets, API keys, subscription credentials, customer data, or private repository content in agent memory unless the user explicitly approves and the environment is trusted.
- Do not bypass access controls, scrape paid content, or expand privileges.