Copilot client parity: Squad works fully only on GitHub Copilot CLI

Problem

Squad currently achieves full feature parity only on the GitHub Copilot CLI (terminal client). The task tool (sub-agent spawning), /delegate (background work), model selection, and background task management all work as designed in the CLI.

In VS Code and other Copilot surfaces, equivalent capabilities exist but may use different tool names (e.g., runSubagent instead of task), have different behavior, or may not support all features Squad relies on.

Key Areas Needing Investigation

Sub-agent spawning: Tool name differences across clients (task → runSubagent)
Background/async agent execution: Workflows differ; CLI has /delegate and /tasks
Model selection parameter support: Per-agent model override may not be exposed
MCP server integration: .ai-team/ discovery and MCP reads
Custom agent file discovery: How agents find .github/agents/ and local charters

Goal

Squad should work across all Copilot surfaces — CLI, VS Code, JetBrains, GitHub.com — with graceful degradation where features aren't available.

Key insight: Squad's architecture is portable (it's just markdown and prompts), but the tool interface assumptions need validation per client.

Success Criteria

Document tested behavior matrix (CLI ✅, VS Code ❓, JetBrains ❓, GitHub.com ❓)
Identify which Squad patterns work on which clients
Define fallback strategies (e.g., what does Squad do if task isn't available?)
Test sub-agent spawning on VS Code with runSubagent
Update squad.agent.md with client compatibility notes

Refined Scope: MCP Integration Documentation (from Brady)

Key insight: Squad members already use MCP tools — the GitHub MCP server is active in this session right now (github-mcp-server-* tools). This isn't future work. It's happening.

What @copilot should deliver

1. Design Doc — How MCP tools integrate with Squad:

How MCP servers become available to squad agents (platform-level, not Squad-level)
How squad.agent.md references MCP tools in routing and spawn prompts
How customers with custom MCP servers (database, internal APIs, etc.) can make those tools available to their squad members
Architecture: what's platform-provided vs. what Squad orchestrates

2. Documentation — For docs/:

How-to guide: "Using MCP Tools with Your Squad"
Examples: GitHub MCP server (already working), custom MCP servers
Pattern: how to mention MCP capabilities in agent charters so agents know what tools they have
Pattern: how to route work to agents based on MCP tool availability

3. Sprint Plan — Concrete work items for full MCP integration:

Document current state (what works today)
Identify gaps (what should work but doesn't)
Customer-facing docs and examples
Any squad.agent.md updates needed

Constraints

All deliverables file in bradygaster/squad only
Focus on documenting what already works, then identify gaps
Don't build new MCP infrastructure — document how to use what's there

cc @copilot

v0.4.0 Sprint Plan: Copilot Client Parity

Verdict: ✅ GO — This is table stakes for Squad's growth. We can't stay CLI-only.

Context

Brady's refined scope nails it: we need to document what works today, identify gaps, and design graceful degradation. We're not rewriting Squad for every client — we're making it adaptive.

Work Items

WI-1: Client Capability Matrix

Owner: Kujan (SDK/Testing Specialist)
Deliverable: Test matrix showing what works on each surface
- CLI (baseline ✅)
- VS Code (via
  unSubagent, ask tool)
- JetBrains (Copilot API surface)
- GitHub.com (web constraints)
Test cases:
- Background mode (mode: "background")
- Model selection (model parameter)
- MCP tool access (file ops, git, web_search)
- Sub-agent spawning ( ask tool)
Acceptance: Markdown table with ✅/⚠️/❌ + notes for each capability × client

WI-2: Graceful Degradation Design

Owner: Verbal (Prompt Engineer)
Deliverable: Fallback strategy doc
- What does Squad do when �ackground mode unavailable? (fall back to sync)
- How do agents handle missing MCP tools? (notify user, suggest CLI)
- When should Squad refuse to start? (critical tools missing)
Acceptance: Decision doc in .ai-team/decisions/ + prompt additions for agent instructions

WI-3: squad.agent.md Client Compatibility Section

Owner: Verbal (Prompt Engineer)
Deliverable: New section in squad.agent.md
- "## Client Compatibility" heading
- Per-client capability summary
- Prompt additions: "If running in VS Code, prefer X over Y..."
Acceptance: Agents have context to adapt behavior based on client

WI-4: User-Facing Documentation

Owner: McManus (Documentation Specialist)
Deliverable: New guide: docs/copilot-clients.md
- "Using Squad across Copilot clients"
- What works where (user-friendly version of WI-1)
- When to use CLI vs. VS Code vs. other clients
- Known limitations + workarounds
Acceptance: Clear, tested user guide ready for squad.dev

WI-5: VS Code Testing & Validation

Owner: Kujan (SDK/Testing Specialist)
Deliverable: Validation that
unSubagent works for squad members
- Test spawning Kujan, McManus, Verbal via VS Code
- Document any quirks (path resolution, template loading)
- Create repro cases if broken
Acceptance: Confirmation Squad agents can spawn in VS Code OR documented blocker for GitHub

Phases

Phase 1: Discovery (WI-1, WI-5 in parallel)

Kujan runs tests, fills capability matrix
Blocks: Phase 2 (need to know what's broken before designing fallbacks)

Phase 2: Design (WI-2)

Verbal designs degradation strategy based on WI-1 findings
Blocks: WI-3 (prompts need design decisions)

Phase 3: Documentation (WI-3, WI-4 in parallel)

Verbal updates squad.agent.md
McManus writes user guide
Both depend on WI-1 + WI-2 completion

What We're NOT Doing

❌ Client-specific builds — No separate VS Code extension, JetBrains plugin, etc.
❌ Feature parity enforcement — If GitHub.com can't do background mode, that's fine. Document it.
❌ Web UI — Not building squad.dev interface for spawning agents.
❌ Polyfills for missing APIs — We adapt, we don't reimplement Copilot's runtime.

Success Criteria

Users know which client to use for their workflow
Squad degrades gracefully when features unavailable
Agents have prompts to adapt to client constraints
No surprises — limitations are documented upfront

Timeline

Target v0.4.0 release. Phase 1 should wrap in 2-3 days. Full plan: ~1 week.

— 🏗️ Keaton (Lead) · Posted by Squad

🏷️ Closing as research parent — implementation issues created

This issue served its purpose: we identified the problem (CLI-only parity) and Keaton posted a sprint plan. But the plan's Phase 1 is literally 'test and find out' — we don't know what's broken yet.

Decomposed into:

#TBD: Spike: Test Squad on VS Code (runSubagent, background mode, MCP tools)
#TBD: Spike: Test Squad on JetBrains + GitHub.com
#TBD: Write graceful degradation design (after spikes complete)
#TBD: Add Client Compatibility section to squad.agent.md
#TBD: Write user-facing docs/copilot-clients.md

Sub-issues will be linked below once created.

— Squad (Coordinator)

Decomposed into sub-issues with VS Code as priority:

• #32 (P0): VS Code
unSubagent\ compatibility
• #33 (P0): File discovery & .ai-team/ access
• #34 (P1): Model selection & background mode
• #35 (P1): Compatibility matrix document
• #36 (P2, deferred v0.5.0): JetBrains + GitHub.com research

VS Code parity target: v0.4.0. Broader surfaces: v0.5.0+.

✅ Issue #10 — Closing Summary

All deliverables for Copilot client parity have shipped:

Sub-issues (all closed)

VS Code: Test and document runSubagent as Squad spawn mechanism #32 — runSubagent API research → Proposal 032a, 032b
VS Code: Test agent file discovery and .ai-team/ access #33 — File discovery spike → Proposal 033a
VS Code: Model selection and background mode parity #34 — Model/background parity → Proposal 034a
Create client parity compatibility matrix document #35 — Client compatibility matrix → docs/scenarios/client-compatibility.md

Shipped artifacts

Client Compatibility Matrix — \docs/scenarios/client-compatibility.md\ (270 lines, full feature-by-platform breakdown)
VS Code How-To Guide — \docs/features/vscode.md\ (user-facing guide: getting started, differences, tips, limitations)
squad.agent.md Client Compatibility section — Platform detection, VS Code spawn adaptations, feature degradation table, SQL tool caveat
VS Code spawn template annotations — Background and sync templates annotated with VS Code equivalents
Cross-platform spawning skill — .ai-team/skills/cross-platform-agent-spawning/SKILL.md\

Key architectural decisions

Platform detection via capability probing (check for \ ask\ vs
unSubagent)
VS Code: multiple
unSubagent\ calls in one turn = parallel (replaces background mode)
Model selection: accept session model in Phase 1, per-agent .agent.md\ frontmatter in Phase 2
SQL tool is CLI-only — cross-platform paths avoid SQL dependency

feat: client parity, VS Code guide, context budget update

fix: apply P0+P1 security audit findings for remote-control

feat: Squad Remote Control - PTY mirror + devtunnel for phone access (#…

Navigation Menu

Repository navigation

Copilot client parity: Squad works fully only on GitHub Copilot CLI #10

Problem

Key Areas Needing Investigation

Goal

Success Criteria

Related

Refined Scope: MCP Integration Documentation (from Brady)

What @copilot should deliver

Constraints

v0.4.0 Sprint Plan: Copilot Client Parity

Context

Work Items

WI-1: Client Capability Matrix

WI-2: Graceful Degradation Design

WI-3: squad.agent.md Client Compatibility Section

WI-4: User-Facing Documentation

WI-5: VS Code Testing & Validation

Phases

What We're NOT Doing

Success Criteria

Timeline

🏷️ Closing as research parent — implementation issues created

✅ Issue #10 — Closing Summary

Sub-issues (all closed)

Shipped artifacts

Key architectural decisions

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Copilot client parity: Squad works fully only on GitHub Copilot CLI #10

Description

Problem

Key Areas Needing Investigation

Goal

Success Criteria

Related

Activity

bradygaster commented on Feb 12, 2026

Refined Scope: MCP Integration Documentation (from Brady)

What @copilot should deliver

Constraints

bradygaster commented on Feb 13, 2026

v0.4.0 Sprint Plan: Copilot Client Parity

Context

Work Items

WI-1: Client Capability Matrix

WI-2: Graceful Degradation Design

WI-3: squad.agent.md Client Compatibility Section

WI-4: User-Facing Documentation

WI-5: VS Code Testing & Validation

Phases

What We're NOT Doing

Success Criteria

Timeline

bradygaster commented on Feb 13, 2026

🏷️ Closing as research parent — implementation issues created

bradygaster commented on Feb 13, 2026

bradygaster commented on Feb 13, 2026

✅ Issue #10 — Closing Summary

Sub-issues (all closed)

Shipped artifacts

Key architectural decisions

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions