Add Default reasoning reset to chat

[bkudiess]

Related to #980

Stack

This is Layer 3 of a three-PR stack:

1. Layer 1, Extract pending gateway request registry #1090 at 79b289e0d7d91e05a6f7da507d3cf25ec9927dfd: PendingRequestRegistry owns request correlation and response classification.
2. Layer 2, Extract thinking-level reconciliation #1094 at 6480e8702f9bebe94cffbc2423bc3e60a3092b55: ThinkingLevelClearReconciler owns reasoning-clear operation, generation, ACK, retry, supersession, and disposal state.
3. This PR at 96549f9766e5e30812482251de75d76d628ea23a: adds the focused Windows Default/reset behavior, provider delegation and cache convergence, architecture closure guards, typed protocol, proof, and tests.

This PR remains draft and must not merge before its dependencies.

Root cause

Choosing any concrete reasoning level, including off, persisted a session ThinkingLevel override. The tray exposed only concrete levels and labeled medium as the default, even though the Gateway protocol supports clearing a session field through SessionPatch.Clear.

After a clear was confirmed, the bridge cache could still replay a stale concrete value during same-provider reload/remount. That stale cache row could overwrite the provider's confirmed canonical null even though the Gateway remained on Default. Separately, a rejected or interrupted clear could leave the reconciler confirmation task faulted after the caller stopped awaiting it.

Behavior

• Adds a localized Default reasoning entry, distinct from off and every concrete level. medium is no longer labeled as the default.
• Threads a typed clear action through the composer, chat root, provider, bridge, and Gateway client.
• Serializes the clear as thinkingLevel: null. It never sends "default" or substitutes "medium".
• Preserves the prior visible canonical value until both the sessions.patch ACK and a current-generation, response-correlated sessions.list snapshot confirm null.
• Represents an ACKed clear as committed but unconfirmed when caller cancellation or confirmation timeout occurs. Bounded reconciliation continues independently.
• Protects thread, operation, client, and connection-generation ownership. Stale callbacks, reconnects, rapid clears, concrete supersession, canceled requests, duplicate responses, and old-client responses cannot clear the wrong thread or roll back newer intent.
• Preserves unrelated fresh session fields while the reasoning field is protected.

Final Layer 3 corrections

1. Confirmed-null cache convergence. OpenClawChatDataProvider keeps a provider-instance, per-thread authority token after an accepted correlated null confirmation. Ordinary bridge/cache rows and non-target correlated rows clone and mask only stale non-null ThinkingLevel; they never mutate bridge-owned objects. Tokens retire only on ordered evidence: ordinary/cache null, row disappearance, connected-authority transition, provider disposal/recreation, the thread's correlated confirmed non-null, or a successful later concrete selection that still owns the exact captured token. A newer clear cannot be retired by an older concrete operation.
2. Fault observation. The provider immediately attaches the existing fault-only observer to ClearOperation.Confirmation. Await behavior, surfaced errors, timeout behavior, and cancellation semantics remain unchanged, while direct rejection and post-ACK disconnect faults cannot escape as matching UnobservedTaskException.

Ownership and compatibility

• OpenClawComposer and OpenClawReactorChatRoot own the Default menu state and accessible checked semantics.
• OpenClawChatRoot and OpenClawChatDataProvider adapt typed user intent and canonical session updates.
• Layer 2 ThinkingLevelClearReconciler remains the sole owner of reasoning-clear lifecycle state. The provider does not duplicate operation maps, retry timers, generation state, or cancellation ownership.
• Layer 1 PendingRequestRegistry remains the sole owner of response correlation. The Gateway client does not add feature-specific request maps or completed-ID stores.
• IChatGatewayBridge and OpenClawGatewayClient own the typed protocol boundary and exact wire payload.
• Null remains the provider-default representation. Windows does not infer, seed, or cache a concrete provider default.
• Older Gateways continue through the existing sessions.patch clear contract. Unknown or unsupported methods surface a truthful error and retain the prior selection.
• Production changes in the final publication commit are confined to OpenClawChatDataProvider; Shared, Gateway, reconciler, root, coordinator, and public API surfaces are unchanged.

Upstream core capability and model-validation work remains separate and out of scope: openclaw/openclaw#99240, openclaw/openclaw#99798, and openclaw/openclaw#109623.

Validation

Validated on exact Layer 3 head 96549f9766e5e30812482251de75d76d628ea23a with OPENCLAW_REPO_ROOT set to the isolated worktree:

Command / suite	Exact-head result
git diff --check	Passed
.\build.ps1	Passed: 5 projects, 0 warnings, 0 errors (69.20s)
dotnet test .\tests\OpenClaw.Shared.Tests\OpenClaw.Shared.Tests.csproj --no-restore	3,428 passed, 32 skipped, 0 failed (33.64s wall)
dotnet test .\tests\OpenClaw.Tray.Tests\OpenClaw.Tray.Tests.csproj --no-restore	2,087 passed, 0 failed (52.2s wall)
dotnet test .\tests\OpenClaw.Connection.Tests\OpenClaw.Connection.Tests.csproj --no-restore	535 passed, 0 failed (21.19s wall)
dotnet test .\tests\OpenClaw.SetupEngine.Tests\OpenClaw.SetupEngine.Tests.csproj --no-restore	703 passed, 0 failed (47.1s wall)
dotnet test .\tests\OpenClaw.WinNode.Cli.Tests\OpenClaw.WinNode.Cli.Tests.csproj --no-restore	126 passed, 0 failed (16.5s wall)
Functional UI	19 passed, 0 failed (9.75s wall)
Native WinUI	117 passed, 0 failed, 0 skipped (112.04s wall)
Axe accessibility	19 passed, 0 failed, 0 skipped (89.33s wall)
Layer 3 protocol/UI/localization/architecture filter	24 passed, 0 failed (19s)
Detached request-routing proof collector	3 passed, 0 failed (8.3s); OPENCLAW_RUN_E2E unset

Focused and stress validation

Focus	Exact-head result
Provider T1-T11 plus concurrency, clone, GC/finalization, and integration-safe provider coverage	5 consecutive runs of 312 tests: 1,560 passed, 0 failed (110.05s)
Layer 2 reconciler/concurrency	10 consecutive runs of 17 tests: 170 passed, 0 failed (28.90s)
Layer 1 registry/routing/protocol	5 consecutive runs of 53 tests: 265 passed, 0 failed (18.19s)
Gateway registry/migration/stale-event guards	5 consecutive runs of 64 tests: 320 passed, 0 failed (15.35s)

The final readiness audit reapproved this exact two-file diff with high confidence and zero findings after the one accepted independent-review issue was fixed and all required validation was rerun.

Real behavior proof

Exact-head automated evidence

Exact-head deterministic tests prove:

• First mount preserves a non-null bridge value unchanged.
• After a correlated canonical null, repeated same-provider LoadAsync/remount reads remain null even while the bridge cache is stale.
• Failed clear semantics retain the prior protected value and start exactly one correlated refresh from a cached null.
• Ordinary stale off is masked; ordinary null releases authority; later concrete canonical evidence is accepted.
• Reconnect authority changes, provider recreation, and row disappearance release provider authority.
• A snapshot correlated to thread A cannot overwrite or release thread B's authority, whether B is non-null or null.
• Successful concrete selection retires only the exact captured authority token; failed concrete selection and older-concrete/newer-clear races retain null authority.
• Bridge-owned SessionInfo objects are not mutated, while unrelated fresh fields remain authoritative.
• Direct rejection and post-ACK timeout followed by disconnect produce no matching unobserved confirmation fault after forced GC/finalization.

The loopback protocol proof retains the typed clear request shape:

{"type":"req","id":"<request-id>","method":"sessions.patch","params":{"key":"agent:main:main","thinkingLevel":null}}

It also asserts the frame does not contain "thinkingLevel":"default" or "thinkingLevel":"medium".

Hosted exact-head proof

• Credential-free C# CodeQL advanced setup completed successfully against the exact fork head: run 31042800679. The advanced-setup gate and C# security jobs passed; the unrelated Actions-language job was skipped by the workflow's C#-only dispatch input.
• The normal Build/Test workflow has no workflow_dispatch entry and pull-request CI selects only PRs targeting main or master. This stacked draft targets Layer 2, so hosted Build/Test and the real reasoning Gateway E2E cannot be dispatched for this exact stack head without changing workflows.

Exact-head media

Not verified / blocked. The current controller does not expose an active interactive screen-capture handle shared with the app. The prior normal-profile attempt on historical head 83386c86ec15546a17aeb0d293bc4e23034c1b3d restored Default but Graphics.CopyFromScreen failed with The handle is invalid; no blank or unsafe frame was uploaded. Exact-head capture was not attempted without a privacy-safe interactive desktop.

The images below are historical UI references only, captured from prior Layer 3 head c4f48b62c7f3729ade8293c77f9a50f3911ecadd with an isolated temporary WinUI test host and synthetic props. They are not current-head proof.

Historical canonical null, with Default checked:

Historical explicit off, with off checked and Default independently available:

Historical state	Image SHA-256	Dimensions
Canonical null	73485452c52ea9d67d64f43ed135ea159cec3617d357270be34673c9f132ee40	960x680
Explicit off	ce79bdf3327a92cb1fc80a08ff63834d83280f2f7eed7d3695f4c62460c556d8	960x680

Real Gateway

Not verified / blocked. Local WSL/setup/real-Gateway proof was intentionally not run because this host has a recorded isolation-cleanup incident. No credentials, providers/models, real chats, or user sessions were used.

SessionsPatchThinkingLevelE2ETests remains the credential-free isolated proof for explicit off -> canonical null -> Tray reconnect convergence with state restoration in finally. The workflow selects it only when the finalized stack targets main, so exact-head execution remains blocked until dependencies land or a verified disposable remote is available.

Residual proof gap: behavior against an older or lossy real Gateway under real network timing remains unverified. Unsupported-method, timeout, cancellation, delayed-response, remount, and race paths are covered deterministically.

Review

• Final readiness reapproval: passed with high confidence and zero findings on exact head 96549f9766e5e30812482251de75d76d628ea23a.
• Gemini 3.1 Pro rubber-duck review: 99% confidence, zero actionable findings.
• GPT-5.6 Sol independent full-diff review: 95% confidence, zero actionable findings.
• Claude Opus 4.8 independent full-diff review: 97% confidence, zero actionable findings.
• One verified review finding was fixed before reapproval: pending concrete reconciler state could project over an active confirmed-null authority token. Final projection now reapplies active non-target authority; the overlapping failed-concrete regression covers it.
• A dispose-test note was rejected because the specified T7 contract is provider-instance isolation/no static leakage, which the test directly proves.
• A concrete-send ordering concern was rejected after tracing production: PatchSessionThinkingLevelAsync completes after outbound SendRawAsync, not after the Gateway response. Gemini re-reviewed that evidence and closed clean at 99%.
• Structured autoreview was attempted with the Codex/Claude panel but failed closed before model invocation because the validated local bundle was 1,528,568 bytes, exceeding its 180,000-byte safety limit. The limit was not bypassed; independent read-only reviewers inspected the complete filesystem diff instead.

ClawSweeper must re-review this exact head after the body and hosted proof links are final. Target gate: zero code/security findings and sufficient proof.

Maintainer-only cap: this draft cannot merge until dependent draft PRs #1090 and #1094 land and the stack is finalized. This PR will remain open and draft.

[clawsweeper]

Codex review: needs maintainer review before merge. Reviewed August 5, 2026, 8:44 PM ET / August 6, 2026, 00:44 UTC.

ClawSweeper review

What this changes

This draft adds a Default reasoning menu action that clears a chat session’s explicit reasoning override and reconciles Gateway-confirmed session state.

Merge readiness

⚠️ Ready for maintainer review - 6 items remain

Keep open. Current main still presents an unset reasoning level as medium and has no user action to clear it; this collaborator-authored draft is an active three-layer stack whose dependencies and current-head proof remain unresolved.

Priority: P2
Reviewed head: 96549f9766e5e30812482251de75d76d628ea23a
Owner decision: Required. See Decision needed.

Review scores

Measure	Result	What it means
Overall readiness	🦐 gold shrimp (3/6)	PR readiness rating was derived from proof quality, review findings, security review, and reviewer confidence.
Proof confidence	🌊 off-meta tidepool	Not applicable: Real behavior proof is not required for maintainer- or bot-authored pull requests.
Patch quality	🦐 gold shrimp (3/6)	No actionable review findings were identified.

Verification

Check	Result	Evidence
Real behavior	Not applicable	Not applicable: Real behavior proof is not required for maintainer- or bot-authored pull requests.
Evidence reviewed	6 items	Current main lacks the requested control: Current main maps an unset reasoning level to medium, labels it “medium (default),” and only dispatches concrete thinking-level values. Gateway protocol already supports a null clear: The existing SessionPatch contract explicitly permits an explicit JSON null for ThinkingLevel, so Default can use an established protocol representation rather than a string sentinel. Draft depends on two open stack layers: Local ancestry confirms the Layer 1 and Layer 2 heads are ancestors of this draft, while GitHub context identifies both dependency pull requests as open.
Findings	None	None.
Security	None	None.

How this fits together

The tray chat composer sends reasoning choices through the chat provider and bridge to a Gateway session patch. Gateway responses and session snapshots then determine the reasoning state shown back in chat.

flowchart LR
  A[Chat reasoning menu] --> B[Chat provider]
  B --> C[Gateway bridge]
  C --> D[Gateway session patch]
  D --> E[Gateway response]
  E --> F[Session snapshot]
  F --> G[Displayed reasoning state]

Loading

Decision needed

Question	Recommendation
Should Windows expose a Default reasoning action now, using the established null session-patch representation, before upstream model-metadata work is complete?	Sponsor the staged Default flow: Land the lower request-tracking and reconciliation layers, then restack this draft and require fresh Gateway and UI proof.

Why: The change adds a user-visible reasoning mode and response-aware session API path across a three-PR stack; acceptance requires maintainer judgment beyond mechanical review.

Before merge

• Resolve merge risk (P1) - The two lower stack pull requests must land before this draft can be rebased and validated as a main-based change.
• Resolve merge risk (P1) - Existing sessions with an unset reasoning override will change from visibly “medium” to “Default”; older-Gateway response and snapshot behavior needs current-head upgrade proof.
• Resolve merge risk (P1) - The PR body marks exact-head UI and real-Gateway proof as blocked, and its screenshots are explicitly historical rather than current-head evidence.
• Resolve merge risk (P1) - The final PR tree could not be independently hydrated in this checkout because promisor fetch and GitHub DNS failed.
• Complete next step (P2) - This collaborator-authored draft is deliberately stacked on two open pull requests and needs product sponsorship, dependency landing, rebase, and fresh proof rather than an automated repair.

Agent review details

Security

None.

Review metrics

Metric	Value	Why it matters
Stacked change size	26 files, +2,426 / -49 lines	The submitted pull-request diff spans protocol, chat UI, CI, localization, and tests, so dependency sequencing matters before merge.
Workflow scope	1 E2E filter changed	The draft adds reasoning Gateway E2E coverage to the setup-connect job, while its current feature-branch base prevents that PR workflow from running.

Root-cause cluster

Relationship: partial_overlap
Canonical: #980
Summary: This draft addresses the missing client-side Default restore path, while the canonical report also tracks unresolved upstream provider metadata and validation behavior.

Members:

• canonical: Reasoning effort can't be restored after selecting "off" on Claude Opus 4.7 (gateway rejects every non-off level with "use off") #980 - The issue documents the user-facing one-way off state that this draft’s Default action is intended to mitigate.
• adjacent_distinct: Extract pending gateway request registry #1090 - It supplies request-correlation infrastructure required by this stack but does not add the Default behavior.
• adjacent_distinct: Extract thinking-level reconciliation #1094 - It owns reasoning-clear reconciliation state but intentionally excludes the Default UI and Gateway protocol behavior.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Merge-risk options

Maintainer options:

1. Restack and prove after dependencies land (recommended)
   After Extract pending gateway request registry #1090 and Extract thinking-level reconciliation #1094 land, rebase onto main and add current-head isolated Gateway and WinUI proof before merging.
2. Defer the feature stack
   Pause the stack if maintainers do not want the Windows client to establish Default reasoning semantics ahead of upstream model validation.

Technical review

Best possible solution:

Approve or defer the Default semantics, then land the dependency layers in order; if approved, restack this draft onto main and prove an explicit off-to-Default restore through an isolated real Gateway plus a current-head WinUI capture.

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

No for the full Gateway behavior in this review. Current main source clearly lacks the Default action, but an exact model and real-Gateway run was unavailable and the PR’s own current-head proof is blocked.

Is this the best way to solve the issue?

Unclear. Using the existing null SessionPatch representation is a narrow fit, but the response-aware reconciliation stack and cross-Gateway behavior need maintainer sponsorship and current-head proof.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 90df4a4af995.

Labels

Label justifications:

• P2: This is a bounded chat-session behavior improvement with a known user recovery benefit but no demonstrated urgent outage.
• merge-risk: 🚨 compatibility: Unset existing reasoning state changes from a displayed medium default to a distinct Default selection and must remain compatible with older Gateways.
• merge-risk: 🚨 session-state: The change writes and reconciles persisted per-session ThinkingLevel state across acknowledgement, snapshot, reconnect, and remount paths.
• merge-risk: 🚨 automation: The draft changes the setup-connect E2E filter, but current hosted PR CI does not run for its feature-branch base.
• rating: 🦐 gold shrimp: Overall readiness is 🦐 gold shrimp; proof is 🌊 off-meta tidepool and patch quality is 🦐 gold shrimp.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Not applicable: Real behavior proof is not required for maintainer- or bot-authored pull requests.

Evidence

What I checked:

• Current main lacks the requested control: Current main maps an unset reasoning level to medium, labels it “medium (default),” and only dispatches concrete thinking-level values. (src/OpenClaw.Tray.WinUI/Chat/OpenClawComposer.cs:102, 90df4a4af995)
• Gateway protocol already supports a null clear: The existing SessionPatch contract explicitly permits an explicit JSON null for ThinkingLevel, so Default can use an established protocol representation rather than a string sentinel. (src/OpenClaw.Shared/GatewayProtocolModels.cs:265, 90df4a4af995)
• Draft depends on two open stack layers: Local ancestry confirms the Layer 1 and Layer 2 heads are ancestors of this draft, while GitHub context identifies both dependency pull requests as open. (96549f9766e5)
• Hosted PR validation cannot cover the feature-branch base: Current CI only triggers pull-request workflows for main or master; the draft changes that workflow’s setup-connect filter but its own feature-branch base prevents normal hosted validation. (.github/workflows/ci.yml:7, 90df4a4af995)
• Current-head source could not be fully rehydrated: The exact draft commit exists locally, but its tree objects are promisor-missing and GitHub DNS failed; the structured autoreview could not construct the branch diff. (96549f9766e5)
• Not shipped or merged: No local release tag or current local branch contains the draft head, consistent with it remaining an open draft. (96549f9766e5)

Likely related people:

• Karen: Recent main history shows ownership of chat UI polish on the composer surface that this draft extends. (role: recent chat-composer contributor; confidence: high; commits: 4d09797c927f; files: src/OpenClaw.Tray.WinUI/Chat/OpenClawComposer.cs)
• Dallin Romney: Recent main history includes session classification work adjacent to the provider and Gateway session state consumed by this change. (role: recent session-path contributor; confidence: medium; commits: 23414509b9d0; files: src/OpenClaw.Tray.WinUI/Chat/OpenClawChatDataProvider.cs, src/OpenClaw.Shared/OpenClawGatewayClient.cs)
• Barbara Kudiess: Prior merged main work covers Gateway connection behavior, and the current stack is centered on Gateway request and session reconciliation boundaries. (role: adjacent Gateway contributor; confidence: medium; commits: bc210e048aa9, b46e91e21b45; files: src/OpenClaw.Shared/OpenClawGatewayClient.cs, src/OpenClaw.Tray.WinUI/Services/Connection/GatewayConnectionManager.cs)

Rank-up moves

Optional improvements that raise the rating; they are not merge blockers.

• Tighten the PR description with what changed, how it was validated, and any remaining risk.

Rating scale

Score	Internal tier	Crab rank	Meaning
6/6	S	🦀 challenger crab	Exceptional readiness
5/6	A	🦞 diamond lobster	Very strong readiness
4/6	B	🐚 platinum hermit	Good normal PR; ordinary maintainer review
3/6	C	🦐 gold shrimp	Useful, but confidence is limited
2/6	D	🦪 silver shellfish	Proof or implementation needs work
1/6	F	🧂 unranked krab	Not merge-ready
N/A	NA	🌊 off-meta tidepool	Rating does not apply

Overall follows the weaker of proof and patch quality.
Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

Workflow

• 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.

History

Review history (24 earlier review cycles; latest 8 shown)

• reviewed 2026-08-04T22:05:08.498Z sha 438df74 :: found issues before merge. :: [P2] Move reasoning-clear lifecycle out of the chat provider | [P2] Centralize gateway response tracking before adding clear paths
• reviewed 2026-08-05T06:27:12.135Z sha 438df74 :: found issues before merge. :: [P2] Extract reasoning-clear lifecycle from the chat provider | [P2] Centralize session response tracking before adding another map
• reviewed 2026-08-05T08:16:21.692Z sha c4f48b6 :: needs maintainer review before merge. :: none
• reviewed 2026-08-05T10:03:09.749Z sha 83386c8 :: needs maintainer review before merge. :: none
• reviewed 2026-08-05T20:13:54.295Z sha 96549f9 :: needs maintainer review before merge. :: none
• reviewed 2026-08-05T20:26:51.779Z sha 96549f9 :: needs changes before merge. :: [P1] Wait for concrete patch acceptance before releasing Default authority
• reviewed 2026-08-05T20:31:06.739Z sha 96549f9 :: needs changes before merge. :: [P1] Wait for concrete patch acceptance before releasing Default authority
• reviewed 2026-08-05T22:46:05.713Z sha 96549f9 :: needs maintainer review before merge. :: none

[bkudiess]

@clawsweeper re-review

[clawsweeper]

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: when the review finishes, ClawSweeper will create the durable review comment if needed or update the existing comment in place.

Re-review progress:

• State: Complete
• Detail: The durable review result and its route handoff completed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/30884507451
• Updated: 2026-08-04T06:35:25.061Z

[bkudiess]

@clawsweeper re-review

[bkudiess]

@clawsweeper re-review

[bkudiess]

@clawsweeper re-review

[clawsweeper]

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: when the review finishes, ClawSweeper will create the durable review comment if needed or update the existing comment in place.

Re-review progress:

• State: Complete
• Detail: The durable review result and its route handoff completed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/30988470738
• Updated: 2026-08-05T08:18:35.773Z

[bkudiess]

@clawsweeper re-review

[clawsweeper]

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: when the review finishes, ClawSweeper will create the durable review comment if needed or update the existing comment in place.

Re-review progress:

• State: Failed
• Detail: Durable publication exhausted its retry budget and was retained for operator dead-letter recovery.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/31049684463
• Updated: 2026-08-05T21:41:24.748Z

[bkudiess]

@clawsweeper re-review