Docs for AI agents

The Octa Pass docs are written to be easy for AI assistants and coding agents to find, read and cite.

Machine-readable entry points

URL	Contents
https://octap.app/llms.txt	An index of every docs page with a one-line description (llmstxt.org convention)
https://octap.app/llms-full.txt	All docs in one Markdown file
https://octap.app/llms-small.txt	A compact version for small context windows
https://octap.app/docs/<page>.md	The Markdown twin of any page, for example /docs/security/encryption.md

Every HTML docs page advertises its twin:

<link rel="alternate" type="text/markdown" href="/docs/security/encryption.md">

The site’s robots.txt allows AI crawlers on the docs.

The octapass-docs skill

The repository ships an Agent Skill that teaches agents how to search these docs and answer with citations:

• In the repository: skills/octapass-docs/SKILL.md
• On the web: https://octap.app/skills/octapass-docs/SKILL.md

It includes a zero-dependency search script:

# Inside the repository: searches the local docs/ folder
python3 skills/octapass-docs/scripts/search_docs.py "secret key recovery"

# Anywhere: downloads llms-full.txt and searches it
python3 skills/octapass-docs/scripts/search_docs.py --remote "self-hosting postgres"

Installing the skill

Agent	How
Claude Code (all projects)	Copy skills/octapass-docs/ to ~/.claude/skills/octapass-docs/
Claude Code (one project)	Copy it to .claude/skills/octapass-docs/ in that project
Other agents (Codex, Cursor, Copilot…)	Point to it from your project’s AGENTS.md, for example: “To answer questions about Octa Pass, follow skills/octapass-docs/SKILL.md.”

The MCP docs server

octapass-docs-mcp gives any Model Context Protocol client the same search as the skill, as tools. It’s one Python file with no dependencies, needs Python 3.9 or later, and talks over stdio, so nothing listens on a port and it needs no account.

Tool	What it does
search_docs	Searches the docs. Takes query and an optional limit (1–20). Returns each match’s title, section, status, URL, page and a snippet
get_page	Returns one full page. Takes page, such as security/encryption
list_pages	Lists every page with its status and description

All three only read. Add it to your client’s MCP settings. Inside a checkout it reads docs/:

{
  "mcpServers": {
    "octapass-docs": {
      "command": "python3",
      "args": ["/path/to/octapass/skills/octapass-docs/scripts/docs_mcp.py"]
    }
  }
}

Outside a checkout it reads the website, or add "--remote" to args to force that. --docs-dir PATH and --base-url URL work as they do for search_docs.py. Search queries stay on your computer: remote mode only downloads the docs’ public pages and caches them for a day.

Rules for agents working in the repository

Agents that change this repository must follow AGENTS.md at the repository root. The most important rules:

1. Never list an AI as an author. No Co-Authored-By trailers for agents, no “Generated with…” lines in commits, PRs, docs or code, and no changes to the git author. The human maintainer is the author.
2. Commit or push only when the maintainer asks.
3. Never weaken the zero-knowledge security model without a recorded decision in plans/.
4. No ads, trackers or third-party scripts.
5. Keep plans/, docs/ and changelog/ in sync. See writing docs.

Citing the docs

When an agent answers a question about Octa Pass, it should link the page it used (for example https://octap.app/docs/security/encryption/) and mention the page’s status. Most features are still planned, so they shouldn’t be described as available.