Tray UI: completed ACP/subagent sessions clutter session list — need archive/hide support

[shanselman]

Summary

The Windows tray app (and likely other UI surfaces) shows a growing list of completed ACP/copilot sessions and native subagents with no way to hide or dismiss them. This makes the session list cluttered and hard to navigate.

Current behavior

• Every sessions_spawn call (ACP codex/copilot sessions, native subagents) adds an entry to the session list in the tray UI
• Completed sessions remain visible indefinitely in the UI
• With frequent cron jobs or coding agent usage, the list quickly grows to dozens of entries with names like "copilot", "subagent", etc.

Expected behavior

• Completed/done sessions should be hidden from the tray UI session list (or collapsible/filterable)
• Ideally, a "hide completed sessions" toggle in the tray UI
• ACP sessions should respect agents.defaults.subagents.archiveAfterMinutes the same way native subagents do (or have an equivalent acp.archiveAfterMinutes setting)

Config context

agents.defaults.subagents.archiveAfterMinutes is set to 60 — this helps with native subagents but ACP sessions don't appear to be affected by it.

Suggested fixes

1. Tray UI: Add a toggle or filter to hide status: done sessions from the session list
2. ACP sessions: Apply archiveAfterMinutes (or equivalent) to ACP session lifecycle so they drop off the list after completion
3. Config knob: A sessions.uiShowCompleted: false or similar setting that controls visibility across all UI surfaces

[clawsweeper]

Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve. Reviewed August 9, 2026, 5:27 AM ET / 09:27 UTC.

Summary
Current main already resolves the tray-list clutter, but the remaining ACP archival and cross-surface policy require an upstream product contract. This collaborator-authored issue should stay open for that explicit decision.

Reproducibility: no. for a current Windows tray clutter bug: source and tests show clean completed sessions are hidden by default. The unverified remaining behavior is an upstream ACP retention policy, not a locally reproducible regression.

Ways to help us reproduce this

• Add a screenshot or short recording showing the behavior.
• Add expected vs actual behavior.
• Include redacted logs or terminal output.

Maintainer decision needed

Question	Recommendation
Should this Windows issue be closed as locally complete and the remaining ACP retention policy tracked in the upstream session-lifecycle work, or should this repository own a specific compatible projection change?	Track upstream lifecycle work: Close the local tray portion as complete and use the upstream session-lifecycle discussion to define archive semantics before any client change.

Why: The tray behavior is implemented, while the remaining request introduces cross-surface lifecycle and configuration semantics that the current gateway protocol does not define.

Next step
A maintainer must select the upstream archive and configuration contract before any safe local implementation can be scoped.

Review details

Best possible solution:

Define archival semantics upstream, including terminal states, retention, transcript queryability, and list filtering, then have Windows consume that contract while retaining the existing local visibility preference.

Do we have a high-confidence way to reproduce the issue?

No for a current Windows tray clutter bug: source and tests show clean completed sessions are hidden by default. The unverified remaining behavior is an upstream ACP retention policy, not a locally reproducible regression.

Is this the best way to solve the issue?

Unclear: a gateway-defined archive lifecycle is safer than a second Windows-only policy, but maintainers must choose the desired cross-surface contract first.

AGENTS.md: found and applied where relevant.

Remaining risk / open question:

• Adding a Windows-only archive timer or shared preference without an upstream retention contract could create divergent lifecycle behavior across clients and transcript handling.

Codex review notes: model internal, reasoning high; reviewed against 582a44713860.

Label changes

Label changes:

• remove clawsweeper:not-repro-on-main: Current issue advisory state no longer selects this label.

Label justifications:

• P3: The remaining work is a low-urgency session-management policy choice after the local clutter problem is already filtered on current main.
• impact:ux-friction: Unarchived completed sessions can still make session navigation noisy, but the tray now provides a persisted local hiding path.

Evidence reviewed

What I checked:

• Current tray behavior: The Sessions page passes its loaded gateway sessions through the visibility filter using the persisted completed-session preference before building rows and channel tabs. (src/OpenClaw.Tray.WinUI/Pages/SessionsPage.xaml.cs:188, 582a44713860)
• Default and regression contract: The preference defaults to false, and current tests verify that clean done/completed sessions are hidden while failed, killed, timeout, running, and aborted sessions remain visible. (src/OpenClaw.Shared/SettingsData.cs:141, 582a44713860)
• Archive-policy boundary: The Windows client requests sessions.list with only an optional agentId and consumes the returned snapshot; the checked-in protocol snapshot exposes no archive-retention or archived-filter request field. (src/OpenClaw.Shared/OpenClawGatewayClient.cs:996, 582a44713860)
• Feature history: The current session presentation was most recently simplified by Dallin Romney in commit cdcb74f; the earlier merged tray visibility implementation was commit b37b1be. (src/OpenClaw.Tray.WinUI/Services/SessionVisibilityFilter.cs:9, cdcb74f44a21)
• Release provenance: The current filtering implementation is newer than v0.6.12: ancestry checks show neither the original visibility commit nor the later session-presentation commit is contained in that release, while the latter is on current main. (cdcb74f44a21)

Likely related people:

• Dallin Romney: Recent current-main commits flattened gateway classification and simplified session presentation. (role: recent session-state contributor; confidence: high; commits: 23414509b9d0, cdcb74f44a21; files: src/OpenClaw.Tray.WinUI/Services/SessionVisibilityFilter.cs, src/OpenClaw.Shared/Sessions/SessionRunState.cs)
• Andy Ye: Authored the merged tray visibility implementation that originally addressed the local hiding portion. (role: introduced local visibility behavior; confidence: medium; commits: b37b1bef0a0c; files: src/OpenClaw.Tray.WinUI/Pages/SessionsPage.xaml.cs, src/OpenClaw.Tray.WinUI/Services/SessionVisibilityFilter.cs)
• Scott Hanselman: Authored the merged follow-up session-filter work and is the collaborator who opened this remaining-policy discussion. (role: recent session-filter contributor and issue author; confidence: medium; commits: 258f83858dc7, ee4817d2c2e7; files: src/OpenClaw.Tray.WinUI/Chat/OpenClawChatDataProvider.cs, src/OpenClaw.Tray.WinUI/Services/SessionVisibilityFilter.cs)

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

[llongley]

I'm getting ramped up on the project and would like to pick this one up as a first issue, so I'm dropping a comment to signal I'm looking at it. I'd have self-assigned to make that clearer, but it looks like I don't have permissions on the repo yet - happy for a maintainer to assign me if that's the norm here.

Before I start I want to make sure I build the right thing. The issue lists three fixes - a tray toggle to hide completed sessions, ACP sessions honoring archiveAfterMinutes, and a global sessions.uiShowCompleted config knob. I'd like to start with just the first one: a "hide completed sessions" toggle on the tray Sessions page that filters the already-loaded session list. The ACP lifecycle piece and the cross-surface config knob feel like separate changes, so my instinct is to land the tray filter first and treat those two as follow-ups. Does that split make sense, or would you rather see them bundled?

A few things I'd want your read on before I start:

1. Should completed sessions be hidden by default (with the toggle to bring them back), or shown by default (with the toggle to hide them)? I'm leaning hidden-by-default, since that's the clutter the issue is about, but it's your call.
2. Should the toggle persist across restarts? I'm assuming yes - a visibility preference that resets every launch would feel broken - but wanted to confirm.
3. Which statuses count as "completed"? I had a look at the gateway repo, and a session's run status is one of running | done | failed | killed | timeout (session-utils.types.ts), where terminal means everything except running. My instinct is to hide only clean done and keep failed/killed/timeout visible, since a run that failed or got killed is probably the one you want to notice. Happy to hide all four terminal statuses instead if you'd rather "completed" mean "any finished run".
4. That same gateway row carries abortedLastRun separately from status, so my plan is to keep any session with that flag set visible even when its status would otherwise hide it. Shout if you'd rather done always win.

Thanks!

[shanselman]

@llongley no one really has permissions, so just send PRs. Your comment was 5D ago, and now there's a PR from another, can you take a look at that and validate?