Red5 AI Skills

AI agent skills for building with Red5 — Red5 Pro (self-hosted) and Red5 Cloud (managed PaaS). Point a skill-aware AI coding assistant (e.g. Claude Code) at this repo and it can answer questions about Red5 setup, streaming protocols, authentication, SDKs, and more without guessing — every technical claim in the primary skill traces back to Red5's own documentation.

Skills in this repo

| Skill | What it covers | |---|---| | red5pro | The primary technical reference: server setup/clustering, streaming protocols (WebRTC/WHIP/WHEP, RTMP/ERTMP, RTSP, HLS), authentication, streaming features (restreaming, transcoding, mixing, recording), Stream Manager 2.0, Red5 Cloud, and client/backend SDKs (Web, Android, iOS, Conference, Backend, Core). | | red5-cloud-pubnub | Setting up Red5 Cloud's built-in PubNub integration — chat, presence, and image/file sharing alongside live video, with copy-pasteable HTML5/iOS/Android code. Also covers migrating/positioning from Agora — see that skill's Guardrails for what is and isn't verified. |

Quick Install With npx (Claude Code, Codex)

npx red5pro-skills

The wizard detects which tool you're using (Claude Code, Codex), lets you pick user- or project-level install, downloads this repo's skills/ directory straight from GitHub, and installs both skills into the right location. Pass --target <name> to skip the prompt or --list-targets to see the full list. Source: bin/cli.js.

Only tools with a folder-based skills convention (SKILL.md + references/ subdirectories) are supported here — Cursor, Windsurf, and GitHub Copilot use a single flat instructions file instead, so use the manual copy method below for those.

Using a skill manually

Copy the skill directory you want (e.g. skills/red5pro/) into wherever your AI tool loads skills from — for Claude Code, typically a project's .claude/skills/ directory or your global ~/.claude/skills/ — or clone this whole repo and point your tool at it. Each skill is self-contained: SKILL.md is the entry point, references/ holds the actual content the skill loads on demand.

How this repo is organized

• AGENTS.md — repository structure, the source-of-truth/freeze-forever rule, naming conventions, and the process for adding a new topic or a whole new skill. Read this before contributing.
• ARCHITECTURE.md — the red5pro skill's internal 4-layer progressive-disclosure design (why content is split the way it is).
• scripts/ — validate-skills.sh (static checks: frontmatter, broken links, leaked local paths, blocklisted terms).
• tests/eval-cases.md — prompt → expected-route eval cases for manually verifying skill behavior; no automated harness yet.
• bin/, src/ — red5pro-skills npx installer sources.

Contributing

See AGENTS.md for the full process. In short: every claim in the red5pro skill must trace back to an actual page in Red5's documentation — no invented API names, endpoints, or config. Run bash scripts/validate-skills.sh before opening a PR; it also runs in CI on every push and pull request.

License

Apache License 2.0.