Reading the Docs

Agents can read this site four ways: two static files, one CLI command, and one MCP tool. Pick the one your harness already has. Every page is also served as Markdown at its URL plus .md (for example https://docs.bitbadges.io/agents/setup.md), and the Copy control at the top right of each page copies that same Markdown or a prompt that points an agent at it.

# Curated index (about 51 KB): start here, then fetch 1-3 pages
curl -sO https://docs.bitbadges.io/llms.txt

# Full-text dump of every page (about 1.9 MB); /for-llms.txt is the same file
curl -sO https://docs.bitbadges.io/llms-full.txt

# One page as Markdown
curl -s https://docs.bitbadges.io/agents/setup.md

# From the CLI (fetched from GitHub, cached 24 hours in ~/.bitbadges/docs-cache.json)
bb dev docs                          # navigable tree of sections
bb dev docs all                      # the full dump
bb dev docs messages/msg-transfer-tokens
bb dev docs approvals                # partial match: first section containing "approvals"
bb dev docs --refresh                # clear the cache

# Builder skills (shorthand for docs builder-skills)
bb dev skills
bb dev skills smart-token

Download the Corpus#

Two files, both regenerated with the docs. Every page of this site carries the same two links in its sidebar under For agents, with the current byte size next to each.

Which one to feed your agent:

• Context-limited agent, or one that can fetch URLs: give it llms.txt. It is much smaller than the full corpus, and every line carries the URL of the page to fetch next. This is the default.
• One-shot paste, or an agent with a large context and no network: give it llms-full.txt. Roughly 400k tokens; paste it whole, attach it as a file, or grep it locally and paste the sections you hit.

Both are plain text with no markup beyond the markdown the pages are written in, so they need no preprocessing.

CLI#

bb dev docs [section] browses the same content offline after the first fetch. Sections nest with slashes (bb dev docs token-standard/approval-criteria/merkle-challenges). A single word with no exact match does partial matching. Reference: Dev commands.

MCP Tool#

fetch_docs({ topic }) runs a keyword search over the live docs export and returns the top matching sections. search_knowledge_base({ query, category }) searches the embedded docs, learnings, recipes, error patterns, and critical rules that ship inside the builder. Both are listed on MCP Builder Tools.

URL Patterns#

Pages follow predictable paths, so an agent can guess a URL without an index.

Every page carries a description: frontmatter line, one # H1, and code before prose. Message names, query names, and CLI command groups use the exact casing from source.

Related#

• Agents
• Dev commands
• Skills