Skip to content

COMMANDS.md

Latest commit

 

History

History
127 lines (87 loc) · 7.15 KB

COMMANDS.md

File metadata and controls

127 lines (87 loc) · 7.15 KB
Raw

Tau Commands

Auth

/login - Start here Pick a provider, enter your credentials, and Tau saves the setup. No env variables, no config hunt.

Models

/models - Pick your model Live model browser. Fetches the real catalog from your provider API, lets you search, filter, and set the active model.

/models                     open the full picker
/models <query>             search active provider
/models openrouter:kimi     search a specific provider
/model kimi-k2-5            set a model directly

Web Search

WebSearch - Firecrawl-hosted web search The web_search tool is hosted with Firecrawl and works across providers. Firecrawl offers 1k searches/month on free trials; just enter your API key.

Setup is one step: /login -> Firecrawl Search -> paste your Firecrawl API key. After that, agents can search current web information automatically when a question needs live or recent data.

Voice

/hey - Start a voice conversation Turns on voice conversation mode. Hold Space to talk, release to send, and Tau shows what it heard before submitting.

/bye - End the voice conversation Turns voice conversation mode off and stops any spoken reply that is still playing.

Session

/tree - Navigate the session graph Move through your conversation history like nodes, so branches and forks stay understandable.

/clone - Clone the session Create a copy of the current session when you want a backup or a clean duplicate to continue from.

/branch - Open a fork Start a fork from the current point in the session without losing the original path.

/resume - Continue later Resume the last useful session or pick an older one when you want to continue where you left off.

Orchestration

/team-mode - Orchestrator with worker agents Multi-provider agent orchestration. One coordinator delegates work to a team of workers and they communicate both vertically (coordinator <-> workers, for task delegation and result handoff) and horizontally (worker <-> worker, for direct collaboration without round-tripping through the coordinator). Each worker can run on a different provider/model, and the orchestrator automatically falls back when a worker fails so the team keeps moving.

Monitoring and Reporting

/usage - Watch provider usage Shows real streaming provider usage as it happens, so you can see provider consumption while working.

/statistics - Review the current session Shows statistics for the active session, including session activity and tool-call details.

/report - Generate a final report Creates a clean content report for the session in Markdown, PDF, or HTML. This is for readable session quality, not usage statistics.

Features

/tools - Toggle optional prebuilt tools Opens an interactive picker for optional Tau prebuilt tools. Basic agent tools stay fixed. Only available in normal power mode — cheap forces every optional tool off and full forces them all on, so /tools is hidden there.

/tools                     open the picker
/tools off AFT             hide AFT tools from the agent
/tools on ProjectWorkflow  enable a tool again
/tools status              print current state

/mode - Switch power mode (cheap / normal / full) One switch for how much machinery Tau loads, with a matching accent color that cross-fades on change.

  • cheap - core tools only. Optional tools, skills, agents, plugins, MCP, and LSP are all off AND hidden from the model (system prompt and listings included); folder configs (.claude/skills, .claude/agents, .mcp.json, plugins) are ignored. Soft bronze accents.
  • normal - default behavior. Your /tools toggles apply; MCP, skills, agents, and LSP load as configured. Standard theme.
  • full - everything on. Every optional tool is enabled regardless of saved /tools toggles. Soft gold accents.
/mode          open the picker (live palette preview)
/mode cheap    minimal footprint
/mode normal   back to default
/mode full     everything on

Saved /tools toggles are never rewritten - cheap/full override them while active, and normal restores them (/tools itself only appears in normal mode). Switching modes changes the tool set and system prompt once, so the prompt cache re-warms on the next message and then stays stable.

/fallback - Recover automatically Automatic recovery when a model fails mid-session. Configure a fallback and keep working through provider outages.

/dangerously-skip-permissions - Skip permission prompts in a trusted sandbox Session-only Bypass Permissions mode. Tau shows a warning before enabling it, permission prompts include the same session option, and /dangerously-skip-permissions off returns to Default mode.

Launch Tau directly in this mode:

tau --dangerously-skip-permissions

/whatsapp - Remote control Tau from WhatsApp Link WhatsApp and control Tau from your phone.

/github - GitHub automation (gh required) GitHub workflows inside Tau, powered by the GitHub CLI.

  • issue - Inspect issues for the current repo, or pass an issue URL to inspect that issue.
  • pr - Inspect pull requests (repo-local or via PR URL) and generate gh-backed actions.
  • wrap - Stage -> commit -> (optional changelog) -> push, with one permission gate before network writes.
  • changelog - Generate/update changelog notes from commit history in a consistent style.
  • triage - Classify issues (labels/status) with explicit confirmation before visible changes.
  • release - Release flow: inspect dirty working tree, check CI/CD workflow status, then tag/publish and list runs.

/safetest - Run a file inside a disposable cloud sandbox Upload one file to a fresh E2B VM, run it there, get a clean report back. The local machine never executes anything. Each run gets its own throwaway sandbox that's destroyed at the end.

Setup is one step: /login -> E2B Security -> pick "Auth login" (opens the E2B dashboard in your browser) or "API key" (just paste). After that, /safetest is ready - no env variables, no extra config.

/pin - Pin a constraint to every prompt Save a sentence (or two) and Tau quietly appends it to the end of every message you send - a persistent reminder the model carries through the whole session without you retyping it. Use it for style rules ("reply in French"), guardrails ("never edit files outside src/"), or task focus ("stay on the auth refactor"). Cache-safe by design: only the dynamic tail of the user message changes, so your provider's prompt cache stays warm and the cost is a few extra tokens per turn.

/learned - Self-learning control hub Tau learns as you work: after a substantial task (or on demand) it proposes one critical, general, reusable lesson - a framework gotcha, a whole class of bug to avoid, a hard-won constraint, or your own preference - for you to Approve / Edit / Skip, then carries approved ones into future sessions and projects. Approve and it's saved and used from the next session, no extra step; lessons are always a single portable principle, never project-specific trivia. Open /learned for a navigable menu: view what it has learned, learn from this session, edit or delete a lesson, or toggle self-learning on/off.