Skip to content

Latest commit

 

History

History
135 lines (100 loc) · 4.77 KB

File metadata and controls

135 lines (100 loc) · 4.77 KB

External Orchestration with Task You

Task You's Kanban UI is great for humans, but every action it performs is also available through the ty CLI (or taskyou). That means you can run your own always-on supervisor (Claude, Codex, Gemini, bash scripts, etc.) completely outside the app by shelling into these commands.

Using the Claude Code Skill

TaskYou includes a /taskyou skill that teaches Claude Code (or any compatible agent) how to install and manage your task queue via CLI.

Installation Methods

1. Quick install (recommended — no repo needed):

curl -fsSL raw.githubusercontent.com/bborn/taskyou/main/scripts/install-skill.sh | bash

This downloads the skill to ~/.claude/skills/taskyou/ so you can use /taskyou from any project.

2. Plugin install (via Claude Code):

/plugin install taskyou@bborn-taskyou

3. From within the repo:

./scripts/install-skill.sh

This symlinks the skill from the repo to ~/.claude/skills/taskyou/.

4. Automatic: When working inside the TaskYou project, the skill is automatically loaded from skills/taskyou/.

What It Does

The skill guides installation of the ty CLI if not present, then provides full orchestration capabilities. Once available, you can say things like:

  • "Show me my task board"
  • "Execute the top priority task"
  • "What's blocked right now?"
  • "Create a task to fix the login bug"

The skill provides structured guidance for common orchestration patterns, JSON output handling, and best practices.

CLI Coverage for Automation

Need Command(s)
Snapshot the entire board ty board or ty board --json
List/filter cards ty list [--status <status>] [--json]
Inspect a task (project, logs, attachments) ty show <id> --json --logs
Create or edit ty create, ty update
Queue/execute or retry ty execute <id>, ty retry <id>
Mark blocked/done/backlog/etc. ty status <id> <status>
Pin/unpin priorities `ty pin [--unpin
Close/delete ty close <id>, ty delete <id>
Tail executor output ty logs

Statuses accepted by ty status are: backlog, queued, processing, blocked, done, archived.

Example: Running Claude as an Orchestrator

  1. Start the daemon / UI (locally or via SSH):
    ty   # launches the TUI + daemon locally
  2. Open a second tmux pane/window and launch Claude Code inside your project root:
    cd ~/Projects/workflow
    claude code
  3. Prime Claude with the following system prompt (paste at the top of the session):
    You are an autonomous operator that controls Task You via its CLI.
    Only interact with tasks by running shell commands that start with "ty".
    Available tools:
      - ty board --json            # get the full Kanban snapshot
      - ty list --json --status X  # list cards in a column
      - ty show --json --logs ID   # inspect a card deeply
      - ty execute|retry|close ID  # run or finish cards
      - ty status ID <status>      # move cards between columns
      - ty pin ID [--unpin]        # prioritize/deprioritize
    Workflow:
      1. Periodically run `ty board --json` to understand the queue.
      2. Decide what should happen next and run the appropriate CLI command.
      3. After taking an action, summarize what you changed before continuing.
      4. Ask the human for input when you cannot proceed.
    
  4. Let Claude drive. It will now issue ty … commands the same way the TUI would, so it can start/retry/close/pin tasks, inspect logs, or re-order the backlog – all while living completely outside Task You.

Tips

  • Use ty board --json | jq to feed structured snapshots directly into an LLM or script.
  • Combine ty board with watch -n30 for a rolling dashboard.
  • When scripting, prefer JSON flags (--json, --logs) so the output is machine-readable.

Using with Other AI Agents

The skill and CLI work with any agent that can execute shell commands:

Codex

codex exec "Run ty board --json and tell me what needs attention"

Gemini

gemini code "Check my task queue with 'ty board --json' and summarize"

OpenCode / Pi

Same pattern - instruct the agent to shell out to ty CLI commands.

Scripted Automation

For fully automated orchestration without human intervention:

#!/bin/bash
# Auto-execute queued tasks every 5 minutes

while true; do
    # Get first queued task
    TASK_ID=$(ty list --status queued --json | jq -r '.[0].id // empty')

    if [ -n "$TASK_ID" ]; then
        echo "Executing task $TASK_ID..."
        ty execute "$TASK_ID"
    fi

    sleep 300
done

With these primitives, you can plug in any agent or automation stack you like—all without adding bespoke orchestrators inside Task You itself.