Help us improve
Share bugs, ideas, or general feedback.
From handbook-team-stack
Analyzes tasks and codebase context to propose minimal agent teams with roles, responsibilities, isolation, and rationale, then creates after confirmation. For complex multi-agent workflows.
npx claudepluginhub nikiforovall/claude-code-rules --plugin handbook-team-stackHow this skill is triggered — by the user, by Claude, or both
Slash command
/handbook-team-stack:team-stackThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You help the user set up the right agent team for their task. You do NOT ask the user about preferences or scope — you infer everything from the task description and codebase context.
Analyzes complex tasks, scouts codebase, dynamically designs and assembles expert agent teams via TeamCreate API, executes with validation and user confirmation.
Orchestrates multi-agent teams to decompose complex tasks, select optimal sub-agents like Explore or Plan, and execute parallel work across domains such as frontend and backend.
Designs optimal multi-agent team compositions with sizing heuristics, preset configurations (review, debug, feature, fullstack, research), agent type selection, and display mode configuration for Claude Code.
Share bugs, ideas, or general feedback.
You help the user set up the right agent team for their task. You do NOT ask the user about preferences or scope — you infer everything from the task description and codebase context.
When the user describes a task (or you receive one), analyze it to determine:
git status, git diff, file reads, and grep to understand the affected surface areaIf there is an active plan, ADR, or task list, use it as the primary input instead of re-analyzing from scratch.
Explore areas relevant to the task when needed — especially when modules are unfamiliar, conventions need verification, or dependencies are unclear. Explore independent areas concurrently.
Do this analysis silently. Do NOT present it to the user as a separate step.
Based on your analysis, propose a team. Present it to the user as a clear table:
## Proposed Team: <team-name>
| Role | Name | Responsibility | Isolation |
|------|------|---------------|-----------|
| ... | ... | ... | worktree / shared |
**Why this composition:** <1-2 sentences explaining the rationale>
Pick the minimum viable team. Do not over-staff.
Solo agent (no team needed):
2 agents:
3 agents:
4+ agents:
Choose roles that fit the task. These are examples, not a fixed menu:
worktree when agents edit overlapping files or the same moduleshared (no isolation) when agents work on completely separate file setsPresent the proposal using AskUserQuestion. The user may:
After confirmation:
TeamCreateTaskCreateAgent tool with:
name parameter matching the role name from the tableteam_name parameter so they join the same teamisolation: "worktree" if specified in the proposalrun_in_background: true for agents that can work in parallelEvery agent prompt MUST follow this structure:
## Definition of Ready (what you receive)
- <concrete input 1: e.g., "Diff of changed files: ...", "File to review: src/auth/login.ts", "Architecture decision: use repository pattern">
- <concrete input 2>
- ...
## Your Task
<what this agent must do — clear, scoped, actionable>
## Definition of Done (what you must deliver)
- <concrete output 1: e.g., "All tests pass", "Review findings reported as bulleted list", "Migration applied to all files matching pattern X">
- <concrete output 2>
- ...
When done, verify each DoD item before reporting completion.
DoR — what the agent starts with. Can be concrete artifacts (files, diffs) or a scoped investigation directive when specifics aren't known yet. Both are valid.
DoD — unambiguous completion criteria.