Skip to content

reference.md

Latest commit

 

History

History
207 lines (161 loc) · 6.48 KB

reference.md

File metadata and controls

207 lines (161 loc) · 6.48 KB
Raw

Reference

Part of the TeamAI CLI Usage Guide.

Command Reference

The full list is in skill-data/core/references/commands.md, or run teamai --help and teamai <command> --help.

Configuration Reference

teamai.yaml (remote team config)

team: my-team
description: Team AI resource repo
repo: https://github.com/your-org/your-repo.git
provider: github
# language: en                # optional. en is English only. Omit it, or set zh, and the recall rule, TodoWrite hint, and merged-MR hint stay bilingual
# scope: ignored if present. teamai init --scope sets the local install

reviewers:
  - reviewer1

packages:
  npm:
    - name: typescript
      version: "*"

sharing:
  rules:
    enforced: [code-review-guide]
  recall:
    enabled: false
  docs:
    localDir: ./.teamai/docs   # default ~/.teamai/docs
  env:
    injectShellProfile: true
  coAuthor:
    enabled: false
  gitExclude:
    enabled: true              # init writes true for a new team
  contributeHint:
    enabled: true
  intervention:
    correctionKeywords: []
  webhooks:
    enabled: true
    endpoints:
      - url: https://example.com/hook
        type: json             # json | feishu | wecom
        events: ["*"]          # session-start, session-stop, skill-use, push, pull, or *
        secret: my-signing-key
        timeout: 5000
        retries: 3

config.yaml (local config)

repo:
  localPath: /path/to/.teamai/team-repo
  remote: https://github.com/your-org/your-repo.git
username: your-name
updatePolicy: auto             # auto, prompt, or skip
scope: workspace               # workspace (default from init) or user; project is read as workspace
projectRoot: /path/to/my-workspace
inheritUserScope: true
coAuthorEnabled: true
gitExcludeEnabled: true
contributeHintEnabled: false
usageReportEnabled: false
codexTrustEnabled: false
toolRoots:
  claude: ~/.claude-work
  codex: ~/.codex-alt

The role, project and tag fields hold this directory's own selection. A workspace also delivers the selections of the repos set up inside it, read at each pull and never written here. See Workspaces with repos inside.

Relocated tool roots (toolRoots)

When Claude Code uses CLAUDE_CONFIG_DIR, Codex uses CODEX_HOME, or Cursor CLI uses CURSOR_DATA_DIR / CURSOR_CONFIG_DIR, init records those directories. The accepted keys are claude, codex, cursor-data and cursor-config.

CLAUDE_CONFIG_DIR=~/.claude-work teamai init
CLAUDE_CONFIG_DIR= teamai init    # clear the record
CODEX_HOME= teamai init
CURSOR_DATA_DIR=~/.cursor-data teamai init
CURSOR_CONFIG_DIR=~/.cursor-config teamai init

The directory has to sit in your home or under ~/.config/. Copilot uses its own COPILOT_HOME.

Cursor CLI moves two directories: CURSOR_DATA_DIR holds the transcripts, CURSOR_CONFIG_DIR the cli-config.json the co-author setting goes into. Its user mcp.json, rules/ and commands/ stay in ~/.cursor whatever the two variables say, so those keep their default paths. With CURSOR_CONFIG_DIR unset the config root is $XDG_CONFIG_HOME/cursor when that variable is set, ~/.cursor otherwise.

Webhook notifications (sharing.webhooks)

Event Fires when
session-start A session starts
session-stop A session ends
skill-use A skill is invoked
push teamai push actually pushes
pull teamai pull actually syncs
* Every event above

Tool arguments and output are not sent. A secret adds X-TeamAI-Signature.

teamai webhook list
teamai webhook test
teamai webhook test --dry-run

Model profiles

teamai models switch is what changes an agent. After that, pull keeps switched agents on the team's catalog.

team:<id> comes from models/models.yaml, and from models/<ns>/models.yaml for an active namespace. It has no key. local:<id> is ~/.teamai/models/models.yaml. Write the prefix when the two share an id.

Team catalog

profiles:
  - id: tokenhub
    name: Tencent TokenHub
    base_url: https://tokenhub.tencentmaas.com
    api_key: ${API_KEY}
    model_groups:
      - protocols: [anthropic, openai-chat-completions]
        models:
          - glm-5.3
          - deepseek-v4-flash

api_key must be ${API_KEY}. base_url must not end in /v1. The first model is the default.

Agent Protocol switch writes
Claude Code anthropic ~/.claude/settings.json
Codex openai-responses ~/.codex/config.toml
OpenCode any opencode.json
CodeBuddy / WorkBuddy openai-chat-completions models.json
Pi any ~/.pi/agent/models.json
OMP any ~/.omp/agent/models.yml
Muse Code openai-responses ~/.config/muse/settings.json, and the key as META_API_KEY in ~/.teamai/models/env.sh, which your shell profile loads

The example has no openai-responses group, so Codex and Muse Code are left alone.

Use a team profile

teamai models list
teamai models switch tokenhub
teamai models switch --agent claude --model deepseek-v4-flash
teamai models switch --dry-run

With no name, it lists every profile and asks. Put the key in an environment variable, not in the command:

teamai models configure tokenhub --from-env TOKENHUB_API_KEY
printf '%s' "$TOKENHUB_API_KEY" | teamai models configure tokenhub --api-key-stdin

Team profiles by namespace

models/<ns>/models.yaml replaces the root profile with the same id. Declare it the same way as env. See Env, hooks and MCP servers by namespace.

# manifest/projects.yaml
projects:
  - id: checkout
    resources:
      models: [checkout]

When the gateway address changes, run teamai models switch team:<id> again. Two namespaces with the same id means models are not updated this time.

Personal profiles

teamai models add my-gateway --name "My gateway" \
  --protocol anthropic,openai-chat-completions \
  --base-url https://gateway.example.com \
  --model glm-5.3,deepseek-v4-flash \
  --from-env MY_GATEWAY_KEY
teamai models switch my-gateway

A personal id cannot reuse a team profile's id.

Switching back

teamai models restore
teamai models restore --agent codex
teamai models remove local:my-gateway

restore puts back only the fields TeamAI changed. An agent you edited yourself is skipped by later switches and restores.