Repository navigation
Backlog
Backlog
Copy link
Description
Part of epic #2438 (Phase 7). This is option B of ADR-003 (#2486) as a bounded experiment, not a shipping commitment (team summary on #2469). Option C (#2487) is the 2.0 path and goes ahead independently. This spike doesn't block it, and it doesn't ship.
Question to answer
Would a host-neutral canonical coordinator in the workspace, with Copilot's .github/agents/squad.agent.md generated from it like every other host's projection, measurably improve correctness or maintainability enough to justify migrating existing repositories? A cleaner directory layout alone doesn't count.
Rules for the spike
- Only disposable workspaces that explicitly opt in, for example behind an experimental flag or an environment switch that
squad initnever sets by default. - No default installer migration, and no migration of live teams' repositories.
- A fixed budget, agreed before starting: suggested 3 days. When it runs out, report what was learned, even if the prototype is incomplete.
- The branch is not merged to
devas shipping behavior. Findings come back as a comment here, and as an ADR-003 update if they change the recommendation.
What to build
- A neutral coordinator source in the workspace. Settle open question 2 of ADR-003: inside
.squad/, served bysquad_state_readand followingteamRoot, or outside it. - Copilot and Claude projections generated from it. Copilot's file has to carry what today's file does: the version stamp, the team-capabilities block
squad castwrites (Richer squad.agent.md generation for AW coordinator awareness #1608), and the customization backup (squad upgrade overwrites squad.agent.md — no mechanism to preserve local customizations #1052).
What to test
Reuse the behavioral scenarios from #2487, then add:
- fresh workspaces, and customized legacy workspaces (an edited
squad.agent.md) - monorepos (the Copilot file at the git root,
.squad/in a subfolder), and remoteteamRoot - repeatable upgrades: running
squad upgradetwice gives no diff - customizations preserved, with clear precedence between user edits, project policy and defaults
- rollback restores the original bytes (harness-adapters §13)
- no Copilot regression: the snapshot guard, and a Copilot run of the same scenarios
Report back with
- Did the scenario results improve over C? Count the defects B avoids that C doesn't.
- Maintainability: how much code and how many tests the single source removes or adds, compared with C.
- Migration risk: what breaks for customized, monorepo and remote workspaces, and how it's detected and recovered.
- A recommendation: ship B later, revise it, or drop it.
🤖 Generated with Claude Code
Metadata
Metadata
Assignees
Labels
No labels
Activity
docs(adr-003): record the team direction and link #2487, #2488
docs(adr-003): gains, risks and scope for the host-neutral coordinator (
docs(adr-003): accept option C (#2501)