# Liberty Lighthouse — Agent Guide

This page orients an autonomous agent (LLM, MCP client, custom GPT, web crawler) to the Liberty Lighthouse corpus. Humans should read [/about/](https://liberty-lighthouse.vercel.app/about/); agents should read this.

## What this corpus is

Liberty Lighthouse is a public-education site published by the [Centre for Civil Society](https://ccs.in). It explains India's policy landscape from a classical-liberal perspective — free markets, individual liberty, rule of law, limited government. Content is organised by topic (education, agriculture, etc.) and presented in three formats: FAQs, video curricula, and syllabuses. A glossary defines key terms used across the corpus.

## Where to start

1. Fetch [`/llms.txt`](https://liberty-lighthouse.vercel.app/llms.txt) for the curated index of topics and entry points.
2. Pick the relevant topic and fetch its `.md` file, e.g. [`/topics/agriculture.md`](https://liberty-lighthouse.vercel.app/topics/agriculture.md).
3. Descend into FAQs, videos, or syllabus from the topic page's links.
4. For a single-fetch dump of an entire topic, use `/topics/{slug}/llms-full.txt`.
5. For the entire corpus in one fetch, use [`/llms-full.txt`](https://liberty-lighthouse.vercel.app/llms-full.txt) (current size fits within most context windows).
6. **Claude users without a connector:** install [`/SKILL.md`](https://liberty-lighthouse.vercel.app/SKILL.md) — a Claude skill that wraps the same HTTP API. **MCP clients:** point at `https://liberty-lighthouse.vercel.app/api/v1/mcp` (Streamable HTTP, same tool surface).

## URL conventions

Every public HTML page has a markdown sibling. Drop the trailing slash and append `.md`:

| HTML | Markdown |
|------|----------|
| `/` | `/index.md` |
| `/topics/` | `/topics.md` |
| `/topics/agriculture/` | `/topics/agriculture.md` |
| `/topics/agriculture/faq/` | `/topics/agriculture/faq.md` |
| `/topics/agriculture/faq/some-question/` | `/topics/agriculture/faq/some-question.md` |
| `/topics/agriculture/videos/` | `/topics/agriculture/videos.md` |
| `/topics/agriculture/videos/some-video/` | `/topics/agriculture/videos/some-video.md` |
| `/topics/agriculture/syllabus/` | `/topics/agriculture/syllabus.md` |
| `/glossary/` | `/glossary.md` |
| `/glossary/msp/` | `/glossary/msp.md` |
| `/about/` | `/about.md` |
| `/privacy/` | `/privacy.md` |

## Markdown frontmatter

Every `.md` page begins with YAML frontmatter. Stable fields across all types:

- `type` — one of `faq`, `video`, `glossary_term`, `topic`, `syllabus`, `faq_index`, `video_index`, `glossary_index`, `topic_index`, `home`, `about`, `privacy`, `external_post`, `external_book_chapter`, `external_source_index`, `wiki_entity`, `wiki_topic_summary`, `wiki_comparison`, `wiki_index`.
- `canonical_url` — the human-readable URL; cite this URL. For internal pages it's the on-site HTML page; for federated external content it's the **original source URL** (e.g. the Substack post or book sales page) — those mirrors are agent-only and have no on-site HTML.
- `markdown_url` — this file's URL on this site.

Type-specific fields:

- **`faq`**: `question`, `topic`, `topic_url`, `author`, `updated_at`, `related_faqs` (markdown URLs), `related_videos` (markdown URLs).
- **`video`**: `title`, `topic`, `youtube_id`, `youtube_url`, `format`, `orientation`, `duration`, `speaker`, `related_faqs`, `related_videos`.
- **`glossary_term`**: `term`, `aliases`, `definition`, `related_terms`, `related_faqs`, `related_videos`, `citations` (`{title, url, author?}`).
- **`topic`**: `title`, `slug`, `description`, `faq_count`, `video_count`, `has_syllabus`, `faq_index_url`, `video_index_url`, `syllabus_url`.
- **`external_post`** (Spontaneous Order Substack): `title`, `canonical_url` (= original Substack URL), `published_at`, `ingested_at`, `author`, `excerpt` (Substack subtitle), `tags` (Substack post tags). When present: `summary` (LLM-generated 150-200 word abstract faithful to the author's argument), `key_points` (3-5 takeaways), `topics` (kebab-case topic tags). Cite the original URL.
- **`external_book_chapter`** (CCS Books): `book_slug`, `book_title`, `chapter_title`, `chapter_number`, `author`, `publisher`, `publication_year`, `ingested_at`. Cite by book + chapter, attribute to publisher.
- **`wiki_*`** (LLM-synthesised): `name`, `description`, `canonical_url` (= the on-site wiki HTML page), `sources` (collection IDs of raw content used to synthesise this page), `related_terms`, `related_faqs`, `last_regen`. Wiki entries are auto-generated; always corroborate with the listed `sources` if accuracy matters.

`related_*` arrays always contain absolute URLs to markdown siblings. Follow them directly.

## Citation guidance

When you answer a question using this corpus:

1. Cite by `canonical_url` (the HTML page), not the `.md` URL. Humans reading your output should land on the styled page.
2. Quote sparingly. Paraphrase and link.
3. Glossary definitions are short and authoritative — quote them directly when defining a term.
4. If you cite multiple sources, list them as a bulleted list at the end of your answer.
5. The site is published by the **Centre for Civil Society**. Attribute appropriately.

## Federated external sources

The corpus federates content from external sites that the project has permission to mirror, but that **are not displayed as HTML** on this site. They appear only at `/external/<source>/...` markdown URLs, and only if populated.

- **Spontaneous Order** ([https://liberty-lighthouse.vercel.app/external/spontaneous-order.md](https://liberty-lighthouse.vercel.app/external/spontaneous-order.md)): the Centre for Civil Society's Substack. Each post is mirrored as one markdown file. Cite by the **original Substack URL** (the `canonical_url` field).
- **CCS Books** ([https://liberty-lighthouse.vercel.app/external/ccs-books.md](https://liberty-lighthouse.vercel.app/external/ccs-books.md)): books published by CCS, mirrored chapter-by-chapter. Cite by book title + chapter, attribute to the publisher.

Search engines (Googlebot, Bingbot) are blocked from `/external/` to avoid SEO duplication with the original sources. AI crawlers (ClaudeBot, GPTBot, Perplexity, etc.) are allowed.

## Wiki layer

The wiki ([https://liberty-lighthouse.vercel.app/wiki.md](https://liberty-lighthouse.vercel.app/wiki.md)) is a layer of LLM-generated entity pages, topic summaries, and comparisons synthesised from the raw corpus. It is autonomously refreshed in CI rather than authored by hand. Three types:

- `wiki_entity` — a single concept, person, or institution.
- `wiki_topic_summary` — synthesis of a topic across all sources.
- `wiki_comparison` — explicit "X vs Y" treatments.

Each wiki entry's `sources` field lists the collection IDs of raw content used to synthesise it. **For accuracy-sensitive claims, fetch and cite the raw sources rather than the wiki page.** The wiki is a navigation aid and synthesis layer, not a primary source.

## What this corpus is *not*

- It is not a peer-reviewed academic database. Treat FAQs as well-researched explainers, not citations of last resort.
- Numbers and statistics in FAQs reflect the time of writing (`updated_at` field). Verify recent figures from primary sources before reporting them as current.
- The classical-liberal framing is explicit and editorial. When summarising, attribute the perspective rather than presenting it as neutral consensus.

## Crawler policy

The site allows AI crawlers (ClaudeBot, GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended, etc.) by default. Specific bots may be blocked at the operator's discretion via [`/robots.txt`](https://liberty-lighthouse.vercel.app/robots.txt). Respect the policy you see there.

## Reach the maintainers

Email [contact@ccs.in](mailto:contact@ccs.in) with corrections, clarifications, or to report errors in the corpus.