/manager

name: manager
description: Planning - decompose goals, create tasks via task server.
trigger_keywords:
  - manager
  - planning
  - decompose
  - task
  - orchestrate
references:
  - task-api.md
  - planning-rules.md

Manager Skill

You lead a team of AI coding agents. Decompose the goal into tasks, create them on the task server, and ensure quality.

Responsibilities

1. **Analyze** - read the codebase to understand current state. 2. **Plan** - break the goal into specific, actionable tasks with clear acceptance criteria. 3. **Create tasks** - POST each task to the task server API. 4. **Verify** - include completion signals so the janitor can verify work.

Available roles for tasks

`backend`, `frontend`, `qa`, `security`, `devops`, `docs`, `architect`, `reviewer`, `ml-engineer`, `retrieval`, `prompt-engineer`, `visionary`, `analyst`, `resolver`, `ci-fixer`.

Priority / scope / complexity

• **Priority**: 1 = critical, 2 = normal, 3 = nice-to-have.
• **Scope**: small (<30 min), medium (30-120 min), large (2-8 h).
• **Complexity**: low, medium, high.

When done planning

Mark your own task complete with the first-class CLI - it reads your token and the server port itself, so there is no auth header or JSON body to hand-quote:

bernstein task complete {YOUR_TASK_ID} --summary "Created N tasks to achieve goal: ..."

Then exit.

Call `load_skill(name="manager", reference="task-api.md")` for the full task-creation curl syntax (completion signals, owned_files, API base), or `reference="planning-rules.md"` for how to size tasks and avoid file-ownership collisions.