VS Code: Model selection and background mode parity #34

Activity

1. bradygaster

   added

   go:yesReady to implementReady to implement

   release:v0.4.0Targeted for v0.4.0Targeted for v0.4.0

   type:spikeResearch/investigation — produces a plan, not codeResearch/investigation — produces a plan, not code

   on Feb 13, 2026
2. bradygaster commented on Feb 13, 2026

   

   bradygaster

   on Feb 13, 2026

   CollaboratorAuthor

   More actions

   Proposal 034a: Model Selection & Background Mode Parity — Findings

   Author: Kujan (Copilot SDK Expert)
   Proposal: team-docs/proposals/034a-model-background-parity.md

   Key Findings

   1. Model Selection

   • runSubagent does NOT accept a model parameter — confirmed
   • Model override routes through custom .agent.md frontmatter (model field)
   • Supports prioritized fallback lists: model: ['Claude Haiku 4.5 (copilot)', 'GPT-5.1-Codex-Mini (copilot)']
   • Requires experimental setting chat.customAgentInSubagent.enabled: true
   • Workaround: Accept session model for v0.4.0 (zero work). Model-tier agent files for v0.5.0

   2. Background/Async Mode

   • No mode: "background" equivalent in VS Code
   • But: Multiple runSubagent calls in one turn run in parallel — equivalent concurrency
   • VS Code "Background Agents" are a different concept (CLI-based worktree sessions, user-initiated, not programmatic)
   • No fire-and-forget (Scribe blocks). Mitigation: batch Scribe as last in parallel group
   • No incremental progress (no read_agent). Results arrive all-at-once

   3. The agent Tool

   • runSubagent = anonymous subagent, inherits session model/tools
   • agent = named custom agent with its own model/tools/instructions
   • Squad should use runSubagent in Phase 1, agent in Phase 2 (when custom agent files exist)

   4. Graceful Degradation

   • No model selection: Accept session model, log intent. Not broken, just more expensive
   • No background mode: Spawn all concurrent agents in single turn. Skip launch table. Skip read_agent
   • Neither tool available: Work inline without delegation

   5. Recommendations

   • Phase 1 (v0.4.0): Accept session model, parallel sync subagents, prompt-level conditional instructions
   • Phase 2 (v0.5.0): Generate model-tier .agent.md files during squad init
   • Phase 3 (v0.6.0+): Per-role agent files if custom agent subagent support stabilizes
   • No code-level abstraction layer needed — prompt-level adaptation is sufficient

   Full analysis in the proposal. Builds on 032b and 032a findings.
3. bradygaster commented on Feb 13, 2026

   

   bradygaster

   on Feb 13, 2026

   CollaboratorAuthor

   More actions

   Spike complete. Proposal 034a (model/background parity) shipped. Key findings: no model param on runSubagent, no background mode — workarounds documented.
4. bradygaster

   closed this as completedon Feb 13, 2026
5. bradygaster

   added 3 commits that reference this issue on Feb 13, 2026

   docs(ai-team): add Strausz as VS Code Extension Expert

   ...

   4407384

   docs(ai-team): merge 3 decisions from inbox and propagate updates

   ...

   29f664d

   feat: project board + v0.4.0 docs (#6)

   ...

   b915015