Repository navigation
discussion: Post-0.9.1 routing regression — community feedback #612
Description
Post-0.9.1: Routing regression reports — are you seeing this?
We're seeing consistent community feedback about routing behavior changes after the 0.9.1 release. Before we investigate deeper, we want to check if this is widespread or isolated to specific environments.
What we're hearing
-
Coordinator doing more work than before — Squad coordinators report that they're handling domain work instead of routing properly to agents. The routing decision seems to be defaulting to the coordinator or not activating agent handoffs correctly.
-
Routing feels broken — Agents aren't being routed to in expected scenarios, or routing seems inconsistent (works in some calls, fails in others).
-
Agent names vanishing during execution — The
ame\ parameter on task tool calls isn't being reflected in the CLI UI. Users see generic task IDs instead of meaningful agent names like "eecom" or "flight".
Timing
These reports started coming in after the 0.9.1 release. We're investigating both Squad changes and potential platform changes in the Copilot CLI.
We're actively investigating
Our team is looking into:
- Task tool name parameter handling in recent CLI versions
- Squad routing logic in post-0.9.1 releases
- Whether this is CLI-side or Squad-side
Your help would be valuable: Are you experiencing any of these behaviors? If so, please share:
- CLI version (\copilot --version)
- OS and environment
- Whether you're using discussions, custom agents, or both
- Specific scenario when it happens (e.g., coordinator handoff, domain work)
We'll post findings here as we investigate. Thanks for helping us debug this collaboratively! 🙏
Activity
chore(squad): log work session — triage, fixes, research
jazz127 commentedon Mar 26, 2026
Hi, i'm very new to squad so i apologise in advance i don't really know the correct terms and can't give a great amount of insight. But as a new user, i wanted to let you know that I started out and setup version 0.9.1 just last week. For the first few days, all the routing to teammembers etc seamed great.
I did start getting rate limited and i noticed the coordinator was picking up on this too and said he would make adjustments that seemed to go against the charter. I also asked for some so i could be to blame. i'm not sure if this was the direct cause but the past day or so, i've been noticing the symptoms you describe - it's very rare to see team members and work get farmed out now etc. Once this happens, all my work seems to not include squad related items like updating the .md files etc.
Hope this helps however if it's just noise and not useful to you, please feel free to delete my comment.
fix: add count-based fallback to archiveDecisions() (#626) (#627)
fix: rename template copies to prevent CLI instruction merging (#613) (…
fix: platform-neutral routing enforcement for VS Code (#613) (#617)
bradygaster commentedon Mar 26, 2026
✅ Triage Complete — Mitigations Merged
Thanks for the detailed report, @jazz127! Your observation about the coordinator adapting behavior under rate limits is exactly on point — that's a key piece of the puzzle.
We've completed a triage cycle and merged 10 fixes to address the routing regression. Here's what we found and fixed:
🎯 Root Cause: Template Duplication (PR #624)
The highest-impact fix was discovering that Squad had multiple copies of \squad.agent.md\ (with .template\ extension variants) scattered across the directory tree. The CLI 1.0.11 discovers *.agent.md\ files at every level up to the git root — these duplicates were being merged into the coordinator's prompt, creating conflicting instructions that caused erratic routing behavior.
Fix: We renamed all template copies to .template\ extensions so the CLI doesn't discover them.
This was likely the #1 cause of routing confusion after 0.9.1.
📋 Other Fixes Merged This Session
Agent names vanishing: This is a Copilot CLI platform issue. The
ame\ parameter sometimes drops from context. We're working with the platform team on this.
Rate-limit degradation: When the platform rate-limits, the model can fall back to lower-tier versions, which impacts routing quality. This is an unsolved platform constraint.
🔄 Next Steps
Try the latest dev builds — the template dedup fix alone should dramatically improve routing stability.
If routing is still erratic, please file a new issue with:
Your input helps us pinpoint whether we're still missing a Squad-side issue or if we need to escalate to the platform team.
Thanks for helping us debug this! 🙏
github-actions commentedon Mar 26, 2026
🏗️ Squad Triage — Flight (Lead)
Issue: #612 — discussion: Post-0.9.1 routing regression — community feedback
Assigned to: FIDO (Quality Owner)
Reason: Issue relates to testing/quality work
@copilot evaluation: 🔴 Not suitable — issue involves work outside the coding agent's capability profile.
Team roster:
squad:flightsquad:proceduressquad:eecomsquad:fidosquad:paosquad:capcomsquad:controlsquad:surgeonsquad:boostersquad:gncsquad:networksquad:retrosquad:incosquad:guidosquad:telemetrysquad:voxsquad:dskysquad:simssquad:handbooksquad:ralphsquad:copilotgithub-actions commentedon Mar 26, 2026
📋 Assigned to Flight (Lead)
Issue: #612 — discussion: Post-0.9.1 routing regression — community feedback
Flight will pick this up in the next Copilot session.
bradygaster commentedon Mar 26, 2026
Closing this — all identified mitigations are merged into dev:
If routing regression symptoms persist on the latest builds, please open a new issue with the specifics requested in our triage comment above. That'll help us isolate whether it's Squad-side or platform-side.
Thanks to @jazz127 for the valuable community feedback! 🙏
— Flight 🏗️