Skip to content

member-guide.md

Latest commit

 

History

History
165 lines (112 loc) · 5.45 KB

member-guide.md

File metadata and controls

165 lines (112 loc) · 5.45 KB
Raw

Member Guide

Part of the TeamAI CLI Usage Guide.

Member Onboarding

npm install -g teamai-cli
cd /path/to/my-workspace
teamai init https://github.com/your-org/your-repo

This machine:

teamai init https://github.com/your-org/your-repo --scope user

Skip the platform token and use your existing SSH key or Git credentials:

teamai init https://gitlab.example.com/yourgroup/yourrepo --provider git

This only affects your machine. push pushes the branch; open the PR/MR on the Git host yourself.

Read-only, no git:

teamai init --http https://your-team-host/api --token <api-key>

An agent access token (sk-) works for one agent only. Pass that agent with --agent <tool>; each agent then keeps its own token. While an agent's current token still works, a new one for the same agent is not used, so its delivered models keep working.

Then enroll each workspace you want synced: run teamai bind-project once there. Binding it to a ClawPro project is optional — add --skip to stay unbound, and no prompt asks again. A worktree inherits its main checkout's choice. The HTTP agent leaves a workspace where this never ran alone, saying so once per session; a workspace enrolled but still pending a project keeps asking in each new chat (and after a pause) until it is bound or skipped, and a non-interactive teamai bind-project lists the account's projects so a setup script can act on them.

Running it again with the same endpoint and key keeps them. Moving to another endpoint or key needs --force, and without it nothing is changed:

$ teamai init --http https://new-host.example/api --token <new-key>
...
✖ The HTTP local agent already uses https://your-team-host/api. Re-run with --force to replace it with https://new-host.example/api.
$ teamai init --http https://new-host.example/api --token <new-key> --force

teamai doctor
teamai list

Day-to-Day Use

Auto-sync

A session start runs teamai pull. Run it yourself when the tool has no such hook.

pull leaves a skill, rule, or agent you changed.

teamai pull
teamai pull --dry-run

Team packages

teamai packages

When the team's packages change, session start asks you to run that. It does not install them.

Excluding skills you don't need

teamai skill exclude add using-superpowers
teamai pull

teamai skill exclude remove using-superpowers
teamai pull

This only affects your machine.

Push local resources

teamai push
teamai push --role pm

A new skill, rule, or agent asks which directory it goes to. The choices come from your role and the active projects. An open pull request is updated in place.

Check status

teamai status

Role management

teamai roles list
teamai roles set hai   # unless the team assigns roles (below)
teamai pull

How roles are defined: Distribution mechanism. A team that assigns roles in manifest/members.yaml decides yours: roles list shows them, roles set is refused, and a change goes through that file — see Assigning roles in the team repo.

Tag subscriptions

teamai tags list
teamai tags subscribe frontend testing
teamai tags unsubscribe testing
teamai pull

Commit Co-Author Attribution

In the team repo's teamai.yaml:

sharing:
  coAuthor:
    enabled: false

false removes the commit trailer. true keeps it. Leave the block out and teamai does not change each tool's own setting.

coAuthorEnabled in ~/.teamai/config.yaml on your machine wins. The next teamai pull writes each tool's config.

Auto-Compact Threshold

In the team repo's teamai.yaml:

sharing:
  autoCompact:
    tokens: 400000

AI tools then compact a conversation at their next compaction check once its context reaches about 400K tokens; a long turn can run a little past it before that check. Each tool takes the number its own way:

Tool Setting written
Claude Code, TClaude settings.json autoCompactWindow = tokens + 33K
CodeBuddy, WorkBuddy settings.json autoCompactWindow = tokens ÷ the tool's mid-turn ratio (0.92, or WorkBuddy's 0.9)
Codex, TCodex config.toml model_auto_compact_token_limit = tokens
DeepSeek Harness cordis.patch.yml row compaction-basic, one ratio per model

A tool still compacts by its model's own window when that is smaller, so a model with a small window is not affected. teamai skips a tool and says why when the tool cannot take the number: an environment variable overrides the setting, a custom CodeBuddy model has no maxInputTokens, or no DeepSeek Harness model fits. Cursor has no such setting. Restart the tool after a pull.

These files are shared by the machine, so one scope decides: the user scope when there is one, else the workspace the pull runs in. A value you set yourself is never overwritten. autoCompactTokens in that scope's config wins; 0 opts out. When the team drops the block, the next teamai pull removes the values teamai wrote.

Keeping Delivered Files Out of Git

sharing:
  gitExclude:
    enabled: true

teamai pull lists the paths it delivered in this clone's .git/info/exclude. It does not change .gitignore. The tools still load the files. git status does not show them.

gitExcludeEnabled in this project's local config wins. The next pull applies it.