Skip to content

Comments

Add Agent Teams support for collaborative multi-agent development#492

Closed
ehartye wants to merge 8 commits intoobra:mainfrom
ehartye:claude/add-worktree-guidance-JMZ7n
Closed

Add Agent Teams support for collaborative multi-agent development#492
ehartye wants to merge 8 commits intoobra:mainfrom
ehartye:claude/add-worktree-guidance-JMZ7n

Conversation

@ehartye
Copy link

@ehartye ehartye commented Feb 17, 2026

Summary

This PR adds comprehensive support for Claude's new Agent Teams feature (Opus 4.6+), enabling collaborative development with direct inter-agent communication alongside the existing subagent-driven approach. Agent Teams allow multiple agents to work in parallel with shared task lists and peer-to-peer coordination, complementing the sequential hub-and-spoke subagent model.

Key Changes

New Skill: team-driven-development

  • SKILL.md: Complete skill documentation covering when to use teams vs subagents, team composition patterns, step-by-step setup and execution, communication patterns, and cost management
  • team-lead-prompt.md: Template for orchestrator role (task coordination, communication hub, conflict resolution)
  • team-implementer-prompt.md: Template for implementer teammates (task claiming, TDD, inter-agent coordination)
  • team-reviewer-prompt.md: Template for reviewer teammates (adversarial review, issue categorization, collaborative feedback)
  • example-auth-feature.md: Complete walkthrough implementing an 8-task authentication feature with 4 agents, showing real message exchanges, coordination patterns, and cost comparison

Documentation

  • docs/analysis-agent-teams.md: Gap analysis comparing current subagent architecture with new agent teams capability, including technical architecture, use cases, and implementation roadmap
  • docs/comparison-agent-teams-vs-subagents.md: Decision guide with quick matrix, detailed comparisons, cost structures, and real-world scenarios to help users choose the right approach

Skill Updates

  • skills/using-git-worktrees/SKILL.md: Added multi-worktree readiness audit section covering git habits that break with parallel worktrees, port/service conflicts, and mitigation strategies
  • skills/writing-plans/SKILL.md: Added multi-feature plans section documenting coordination manifest structure for multi-feature initiatives
  • skills/brainstorming/SKILL.md: Added scope check for single vs multi-feature detection
  • skills/finishing-a-development-branch/SKILL.md: Added multi-feature completion section
  • skills/executing-plans/SKILL.md: Added multi-feature context section
  • skills/subagent-driven-development/SKILL.md: Added multi-feature context section
  • README.md: Updated to reference new team-driven-development skill

Implementation Details

  • Coexistence model: Both subagent-driven-development and team-driven-development remain available; users choose based on task requirements and budget
  • Explicit opt-in: Agent teams require CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 environment variable; disabled by default
  • Cost transparency: Clear guidance that teams cost 2-4x more than subagents but provide speed and collaboration benefits
  • Team flexibility: Unlike fixed subagent structure, teams support customizable roles (frontend/backend specialists, security experts, etc.) and sizes (3-6 agents)
  • No breaking changes: All modifications are additive; existing skills and workflows remain unchanged

Use Cases

Agent Teams excel for:

  • Complex features requiring backend/frontend coordination
  • Security-critical work benefiting from adversarial review
  • Emergent dependencies that need negotiation during work
  • Speed-critical projects where wall-clock time matters more than token cost
  • Research/exploration phases requiring multiple perspectives

https://claude.ai/code/session_01KRPBLUKhMYihnYkiWEtsap

Summary by CodeRabbit

  • New Features

    • Introduced Team-Driven Development as an alternative development approach for collaborative coordination.
    • Added multi-feature workflow support with orchestration across development branches.
  • Documentation

    • Added comprehensive Agent Teams architecture and implementation guides.
    • Provided comparison framework between Team-Driven Development and Subagent approaches.
    • Included team member role templates and authentication feature example walkthrough.
    • Added multi-feature coordination context to existing brainstorming and development workflows.

Copilot AI and others added 8 commits February 16, 2026 22:18
Co-authored-by: ehartye <1032978+ehartye@users.noreply.github.com>
Co-authored-by: ehartye <1032978+ehartye@users.noreply.github.com>
Co-authored-by: ehartye <1032978+ehartye@users.noreply.github.com>
Co-authored-by: ehartye <1032978+ehartye@users.noreply.github.com>
Add multi-feature detection to brainstorming scoping phase that prompts
users to focus on one feature (recommended) or proceed with all (higher
risk). When proceeding with multiple features, each feature and shared
dependency gets its own worktree, plan, and agent — preserving single-
agent-per-feature context isolation.

Key additions across 7 skill files:
- brainstorming: scope check detects multi-feature requests, presents options
- writing-plans: coordination manifest with execution order and dependency graph
- using-git-worktrees: multi-worktree creation and dependency distribution
- team-driven-development: orchestrator sequences team deployments per manifest
- subagent-driven-development: multi-feature context awareness
- executing-plans: per-worktree scoping in multi-feature mode
- finishing-a-development-branch: orchestrator integration merge order

The existing single-feature path is fully preserved and unchanged.

https://claude.ai/code/session_01KRPBLUKhMYihnYkiWEtsap
Before creating multiple worktrees, audit the project for issues that
break parallel development: hardcoded ports, shared databases, missing
.env files, platform-specific needs (Salesforce scratch orgs, Python
venvs), git habits like `git add .` and `git stash` that are global
across worktrees, and per-ecosystem dependency installation requirements.

Key additions:
- using-git-worktrees: full "Multi-Worktree Readiness Audit" section
  covering git habits, port/service conflicts, platform-specific
  isolation (SFDX, mobile), dependency installation per ecosystem,
  environment files, and a readiness report template
- brainstorming: multi-feature mode now starts with the readiness audit
  as step 1, with remediation feeding into the shared dependency plan
- writing-plans: coordination manifest gains Phase 0 (Environment
  Readiness) for remediation tasks before any worktree creation

https://claude.ai/code/session_01KRPBLUKhMYihnYkiWEtsap
Copilot AI review requested due to automatic review settings February 17, 2026 00:40
@coderabbitai
Copy link

coderabbitai bot commented Feb 17, 2026

Caution

Review failed

The pull request is closed.

📝 Walkthrough

Walkthrough

This PR introduces comprehensive documentation for a new Agent Teams feature alongside existing subagent-driven development. It includes updated README guidance, new analysis and comparison documentation, a new "team-driven-development" skill with role-specific prompt templates, an end-to-end example, and updates to existing skills to accommodate multi-feature team orchestration contexts.

Changes

Cohort / File(s) Summary
README updates
README.md
Added "team-driven-development" as an activation mode alongside subagent-driven-development and executing-plans, with description of collaborative agent teams and inter-agent communication capabilities.
Agent Teams analysis & comparison
docs/analysis-agent-teams.md, docs/comparison-agent-teams-vs-subagents.md
New comprehensive documentation comparing Agent Teams with subagent approaches, including architecture diagrams, decision matrices, cost-benefit analysis, and scenario-based recommendations for when to use each approach.
Implementation guidance
docs/IMPLEMENTATION-SUMMARY.md
Comprehensive template and knowledge base for Agent Teams feature, including design decisions (coexistence, opt-in enablement), usage patterns, testing strategy, cost implications, and success metrics.
Team-driven-development skill
skills/team-driven-development/SKILL.md, skills/team-driven-development/example-auth-feature.md
New skill documentation describing the team-driven development workflow, team composition (lead, implementer, reviewer, researcher roles), communication patterns, multi-feature orchestration, cost estimation, and a detailed end-to-end authentication feature example with simulated execution timeline.
Team role prompt templates
skills/team-driven-development/team-lead-prompt.md, skills/team-driven-development/team-implementer-prompt.md, skills/team-driven-development/team-reviewer-prompt.md
Three new prompt templates defining role-specific workflows: team lead for orchestration and task coordination, implementer for feature development and task claiming, and reviewer for code review and quality gates with structured feedback categories.
Multi-feature context in brainstorming
skills/brainstorming/SKILL.md
Added multi-feature detection workflow with scope-check decision point, user selection for single vs. multi-feature paths, and detailed guidance for multi-feature handling including shared dependency identification and per-feature approvals.
Multi-feature context in execution skills
skills/executing-plans/SKILL.md, skills/finishing-a-development-branch/SKILL.md, skills/subagent-driven-development/SKILL.md, skills/writing-plans/SKILL.md, skills/using-git-worktrees/SKILL.md
Updated existing skills with multi-feature context sections describing orchestrator-driven execution, per-worktree isolation, shared dependency handling, coordination manifest creation, completion sequencing, readiness audits, and integration patterns.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related issues

Possibly related PRs

Poem

🐰 Behold, the teams now hop together,
With leads and coders, linked by tether,
Shared tasks and prompts, a coordinated dance,
Where agents collaborate at every glance! 🤝✨

✨ Finishing touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Post copyable unit tests in a comment

Tip

Issue Planner is now in beta. Read the docs and try it out! Share your feedback on Discord.


Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands and usage tips.

Copy link

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds documentation-first support for Claude Code “Agent Teams” as a parallel collaboration workflow to the existing subagent model, plus multi-feature orchestration guidance across existing planning/execution skills.

Changes:

  • Introduces new team-driven-development skill with prompt templates and an end-to-end example walkthrough.
  • Adds docs comparing teams vs subagents and an analysis/implementation summary for the feature.
  • Extends several existing skills to support “multi-feature” planning/execution patterns (coordination manifest, multi-worktree readiness audit, per-worktree completion/integration).

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 6 comments.

Show a summary per file
File Description
skills/writing-plans/SKILL.md Adds multi-feature plan/manifest structure and multi-feature execution handoff guidance.
skills/using-git-worktrees/SKILL.md Adds multi-worktree readiness audit and multi-feature worktree/distribution lifecycle guidance.
skills/team-driven-development/SKILL.md New skill defining when/how to run collaborative agent teams, including multi-feature orchestration and cost guidance.
skills/team-driven-development/team-lead-prompt.md New lead prompt template for orchestration, coordination, and escalation patterns.
skills/team-driven-development/team-implementer-prompt.md New implementer prompt template for task claiming, implementation, and review workflow.
skills/team-driven-development/team-reviewer-prompt.md New reviewer prompt template for structured review feedback and escalation patterns.
skills/team-driven-development/example-auth-feature.md Full example of team-driven auth feature execution with message exchanges and cost comparison.
skills/subagent-driven-development/SKILL.md Adds guidance for operating within a multi-feature orchestration context (single plan/worktree scope).
skills/finishing-a-development-branch/SKILL.md Adds multi-feature completion guidance: per-worktree completion and orchestrator integration merges.
skills/executing-plans/SKILL.md Adds multi-feature context constraints and orchestrator/agent responsibilities.
skills/brainstorming/SKILL.md Adds explicit multi-feature detection step and downstream orchestration guidance.
docs/comparison-agent-teams-vs-subagents.md New decision guide comparing approaches, including cost and hybrid usage scenarios.
docs/analysis-agent-teams.md New gap analysis and roadmap doc for adding teams support.
docs/IMPLEMENTATION-SUMMARY.md New implementation summary document for the PR deliverables and design decisions.
README.md Updates workflow and skill list to reference the new team-driven-development skill.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

````markdown
# [Initiative Name] Coordination Manifest

> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans or superpowers:subagent-driven-development per worktree. The orchestrator sequences deployments per this manifest.
Copy link

Copilot AI Feb 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The coordination manifest template calls out executing-plans/subagent-driven-development as the only per-worktree execution skills, but this PR introduces team-driven-development as an additional execution option. Consider updating this line to include team-driven-development (or explicitly state why teams are excluded in multi-feature manifests) to avoid conflicting guidance.

Suggested change
> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans or superpowers:subagent-driven-development per worktree. The orchestrator sequences deployments per this manifest.
> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans, superpowers:subagent-driven-development, or superpowers:team-driven-development per worktree. The orchestrator sequences deployments per this manifest.

Copilot uses AI. Check for mistakes.
- Let agents claim same task (race condition)
- Skip the shared task list (how do agents coordinate?)
- Ignore messages from teammates (breaks collaboration)
- Mix team and subagent approaches in same workflow
Copy link

Copilot AI Feb 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This 'Never' rule says not to mix team and subagent approaches, but the comparison guide explicitly recommends a hybrid approach (teams for coordinated phases, subagents for independent phases). Align these docs by clarifying the intended constraint (e.g., "don’t mix within the same task/worktree") or updating/removing the blanket prohibition.

Suggested change
- Mix team and subagent approaches in same workflow
- Mix team and subagent approaches within the same task/worktree (hybrid across phases is fine; keep each task on one approach)

Copilot uses AI. Check for mistakes.
Comment on lines +113 to +115
- `teammate-implementer-prompt.md` - Team-aware implementer
- `teammate-reviewer-prompt.md` - Collaborative reviewer
- `teammate-researcher-prompt.md` - Exploration and options
Copy link

Copilot AI Feb 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

These prompt filenames don’t exist in the repo (teammate-implementer-prompt.md, teammate-reviewer-prompt.md, teammate-researcher-prompt.md). Update the list to reference the actual templates under skills/team-driven-development/ (team-implementer-prompt.md, team-reviewer-prompt.md, etc.) and clarify that a researcher template is not currently provided if that’s intentional.

Suggested change
- `teammate-implementer-prompt.md` - Team-aware implementer
- `teammate-reviewer-prompt.md` - Collaborative reviewer
- `teammate-researcher-prompt.md` - Exploration and options
- `team-implementer-prompt.md` - Team-aware implementer
- `team-reviewer-prompt.md` - Collaborative reviewer
- *(Researcher role currently has no dedicated prompt template; use the team lead or implementer prompt as a base, or derive a custom researcher prompt as needed.)*

Copilot uses AI. Check for mistakes.
Comment on lines +133 to +135
No modifications to existing skills or workflows:
- New skill added alongside existing ones
- Existing subagent workflows unchanged
Copy link

Copilot AI Feb 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This section claims 'No modifications to existing skills or workflows', but this PR updates several existing skills (e.g., writing-plans, using-git-worktrees, executing-plans, etc.). Reword to reflect the actual scope (e.g., 'no breaking changes' / 'additive changes') so the summary isn’t misleading.

Suggested change
No modifications to existing skills or workflows:
- New skill added alongside existing ones
- Existing subagent workflows unchanged
Additive, backward-compatible updates to existing skills and workflows:
- New skill added alongside existing ones
- Existing subagent workflows preserved

Copilot uses AI. Check for mistakes.

### Single-Feature Handoff

**"Plan complete and saved to `docs/plans/<filename>.md`. Two execution options:**
Copy link

Copilot AI Feb 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Markdown formatting: this bolded quoted handoff line starts with ** but never closes it, which will cause the rest of the section to render as bold. Close the bold marker (and optionally drop the surrounding quotes) so only the intended sentence is emphasized.

Suggested change
**"Plan complete and saved to `docs/plans/<filename>.md`. Two execution options:**
**Plan complete and saved to `docs/plans/<filename>.md`. Two execution options:**

Copilot uses AI. Check for mistakes.

When a coordination manifest exists:

**"Multi-feature plan complete. Manifest and per-feature plans saved to `docs/plans/`. Execution options:**
Copy link

Copilot AI Feb 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Markdown formatting: this bolded quoted handoff line starts with ** but never closes it, so subsequent content may render incorrectly. Close the bold marker (and optionally remove the surrounding quotes) to keep formatting scoped to this sentence.

Copilot uses AI. Check for mistakes.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants