Agent Skills

TiRTC provides a set of skill packages for AI agents.

These skills are not meant to repeat the public docs. Their job is to turn common workflows into instructions that even a low-context agent can execute reliably. External users usually follow one of these two paths:

• read the public docs first, then ask AI to execute the workflow
• install the skill directly and hand the task to AI with a simple prompt

Get and Install

The current skill repository is:

• tangeai/tirtc-skills

Clone the repository locally first:

git clone git@github.com:tangeai/tirtc-skills.git
cd tirtc-skills

Then install the required skill directory into your local skills location according to the tool used by your agent.

The repository README already includes installation instructions for Claude Code and Codex. Follow the repository instructions directly.

Skill List

tirtc-cli-token-issue

This skill generates a connection token for a specific peer_id.

• Typical scenarios: local debugging and temporarily generating a token for a device
• The output can include both the token and the QR code PNG path

Prepare these environment variables before use:

export TIRTC_CONN_ACCESS_ID="<ACCESS_ID>"
export TIRTC_CONN_SECRET_KEY="<SECRET_KEY>"

You can use a prompt like this directly:

Generate a token for <peer_id>

If you want to specify the skill explicitly, you can also write:

Use tirtc-cli-token-issue to generate a token for <peer_id>

An example of the agent's final output:

⏺ Issue succeeded:

  peer_id: TESTAAAAAAAA

  token:
  v1.eyJzdWIi...o

  QR code PNG:
  /Users/.../token-TESTAAAAAAAA-TESTAAAAAAAA-1775745043854.png

  Do you want to open the QR code preview now?

tirtc-cli-server-session

This skill covers the CLI workflow after token generation, including:

• starting a local upstream endpoint for client-side integration
• sending media from a local MP4 file
• sending commands on an existing session
• inspecting and replying to remote commands
• sending stream messages
• exporting logs and reports
• stopping host sessions on the current machine

It does not issue tokens. If you still need a token, use tirtc-cli-token-issue first.

Before starting a server, the user should at least prepare:

• license
• a local MP4 path
• audio_stream_id
• video_stream_id

Among these, license and the local MP4 path must be stated explicitly. They are the two most commonly missed inputs, but they are required by the current CLI contract.

For Android integration, the usual defaults are:

• audio_stream_id = 10
• video_stream_id = 11

You can use prompts like:

Start a local upstream endpoint for client-side integration. The license is XXX, the MP4 path is PATH, audio_stream_id=10, and video_stream_id=11.

Or:

Use this sessionId to send a command request. The command ID is 7953 and the payload is time?

Or:

Stop all TiRTC host sessions on the current machine.

If you want to specify the skill explicitly, you can also write:

Use tirtc-cli-server-session to start a local upstream endpoint for client-side integration. The license is XXX, the MP4 path is PATH, audio_stream_id=10, and video_stream_id=11.

The default behavior of this skill is:

• refresh tirtc-devtools-cli@latest before execution
• call the CLI's own init command to generate the template first, then fill the minimal server.toml
• call service start directly without requiring manual media assets prepare
• return sessionId after startup succeeds
• use explicit --session <SESSION_ID> for all follow-up session actions