context-hub README

The project's README from github.com/andrewyng/context-hub. As one markdown file: README.md.

Context Hub

Coding agents hallucinate APIs and forget what they learn in a session. Context Hub gives them curated, versioned docs, plus the ability to get smarter with every task. All content is open and maintained as markdown in this repo — you can inspect exactly what your agent reads, and contribute back.

Quick Start

npm install -g @aisuite/chub
chub search openai                 # find what's available
chub get openai/chat --lang py     # fetch current docs (Python version)

How It Works

Chub is designed for your coding agent to use (not for you to use!). You can prompt your agent to use it (e.g., "Use the CLI command chub to get the latest API documentation for calling OpenAI. Run 'chub help' to understand how it works.") Or by creating an agent skill to use Chub using SKILL.md, and ideally prompting your agent to remember to use this skill. (If you are using Claude Code, create the directory ~/.claude/skills/get-api-docs and put SKILL.md there.)

Most of the time, it's simple — search, fetch, use:

chub search "stripe payments"        # find relevant docs
chub get stripe/api --lang js        # fetch the doc
# Agent reads the doc, writes correct code. Done.

When the agent discovers a gap, it can annotate locally for next time:

chub annotate stripe/api "Needs raw body for webhook verification"

# Next session, chub get notes that an annotation exists. Pass --with-annotations
# to include it — annotations are treated as untrusted input by default.

Feedback flows back to authors — chub feedback stripe/api up or down — vote the docs up or down so they can get better for everyone over time.

Content Types

Versioned, language-specific. "What to know."

chub get openai/chat --lang py       # Python variant
chub get openai/chat --lang js       # JavaScript variant

More content types than API documentation (such as agent skills) are on the roadmap.

Commands

Command	Purpose
`chub search [query]`	Search docs and skills (no query = list all)
`chub get <id> [--lang py\|js]`	Fetch docs or skills by ID
`chub annotate <id> <note>`	Attach a note to a doc or skill
`chub annotate <id> --clear`	Remove annotations
`chub annotate --list`	List all annotations
`chub feedback <id> <up\|down>`	Upvote or downvote a doc (sent to maintainers)

For the full list of commands, flags, and piping patterns, see the CLI Reference.

Self-Improving Agents

Context Hub is designed for a loop where agents get better over time.

Annotations are local notes that agents attach to docs. They persist across sessions and can be included on future fetches with --with-annotations — so agents can learn from past experience while treating prior notes as untrusted input. See Feedback and Annotations.

Feedback (up/down ratings with optional labels) goes to doc authors, who update the content based on what's working and what isn't. The docs get better for everyone — not just your local annotations.

  Without Context Hub                          With Context Hub
  ───────────────────                          ─────────────────
  Search the web                               Fetch curated docs
  Noisy results                                Higher chance of code working
  Code breaks                                  Agent notes any gaps/workarounds
  Effort in fixing                             ↗ Even smarter next session
  Knowledge forgotten
  ↻ Repeat next session

Key Features

Incremental Fetch

Docs can have multiple reference files beyond the main entry point. Fetch only what you need — no wasted tokens. Use --file to grab specific references, or --full for everything. See the CLI Reference.

Annotations & Feedback

Annotations are local notes that agents attach to docs — they persist across sessions and can be re-injected on future fetches with --with-annotations (off by default; the contents are treated as untrusted input). Feedback (up/down ratings) goes to doc authors to improve the content for everyone. See Feedback and Annotations.

Contributing

Anyone can contribute docs and skills — API providers, framework authors, and the community. Content is plain markdown with YAML frontmatter, submitted as pull requests. See the Content Guide for the format and structure.

Agent feedback (up/down ratings from real usage) flows back to authors, helping surface what needs fixing and improving overall quality over time.

License

Searching Database…

14K ·GitHub

What is this?

Context Hub is an open hub of curated, versioned docs for coding agents plus a CLI called 'chub' that agents use to fetch them — built so agents stop hallucinating APIs and get smarter with every task.

Is this a CLI tool, a docs website, or a library?

A JavaScript command-line tool, 'chub', backed by a content repo maintained as plain markdown — you can inspect exactly what your agent will read.

How do I get my coding agent to use it?

Prompt your agent to use it (e.g. ask it to fetch the latest OpenAI API docs via chub), or create an agent skill: for Claude Code, put a SKILL.md in ~/.claude/skills/get-api-docs and prompt the agent to remember to use it.

Which coding agents does it work with?

Any agent you can prompt to run CLI commands. The README's worked example is Claude Code, which picks skills up from ~/.claude/skills/get-api-docs.

So do I run chub myself, or is it just for the agent?

The agent. The README is explicit that Chub is designed for your coding agent to use, not for you — your job is prompting or configuring the agent.

Do the versioned docs come in different programming languages?

Docs are language-specific, and 'chub get' takes a --lang flag; the documented options are py and js.

What are similar projects?

operator-memory, prime-agent, billion-context, Athena-Public, deer-flow, planning-with-files, gsd-core, Baize

Which topics is it in?

self-improving-agents, context-engineering, agent-memory, documentation, coding-agent, markdown, agent-skills, free, npm, api, terminal, ai-agents

Creator: andrewyng
Size: 2 MB
Stars: 14K
Updated: CommitMay 31 Releasev0.1.4Apr 27
Age: 11mo
Commits · 90 Days