fix: use context-aware reserveTokensFloor in overflow recovery hint by tanshanshan · Pull Request #84399 · openclaw/openclaw

tanshanshan · 2026-05-20T02:24:29Z

Summary

Replace the hardcoded 20000 reserveTokensFloor suggestion in context-overflow recovery messages with dynamic tier-based values based on the model context window size.

Problem

When context limit is exceeded, the error message suggests setting agents.defaults.compaction.reserveTokensFloor to 20000. However, 20000 is too low for models with larger context windows (200k, 1M tokens), causing users to still experience context overflow errors even after applying the recommended config.

Solution

Use context-aware tiered values for the recommended reserveTokensFloor:

100000 for context windows >= 1M tokens
50000 for context windows >= 200k tokens
35000 for context windows >= 100k tokens
20000 for smaller/undefined context windows (default)

Uses session contextTokens as fallback when model metadata is unavailable, with model metadata taking precedence over stale session entries.

Changes

Add computeContextAwareReserveTokensFloor() function with tier logic
Add resolveContextWindowForCompactionHint() to resolve context window with session fallback
Add buildContextOverflowResetHint() helper function
Update buildContextOverflowRecoveryText() to use dynamic hint
Output plain numeric values (e.g. 50000) instead of locale-formatted strings, matching the integer config schema
Add comprehensive unit tests for tier boundaries, fallback behavior, and model-metadata priority

Verification

OPENCLAW_VITEST_MAX_WORKERS=1 pnpm test src/auto-reply/reply/agent-runner-execution.test.ts -t "computeContextAwareReserveTokensFloor|buildContextOverflowRecoveryText" --no-file-parallelism

16 tests passed (7 direct tier tests + 9 integration tests).

Real behavior proof

Behavior addressed: Context overflow recovery hint suggests hardcoded 20000 for reserveTokensFloor regardless of model context window size; after this fix, the hint uses context-aware tiers (100000/50000/35000/20000).

Real environment tested: Local OpenClaw dev setup, verified via pnpm dev gateway with the patched code, invoking buildContextOverflowRecoveryText with different context window sizes.

Exact steps or command run after this patch:

Start gateway with pnpm dev
Call buildContextOverflowRecoveryText with different model context window sizes (1M, 200k, 100k, 32k, undefined)
Observe the recovery hint message for each tier

Evidence after fix:

Terminal output from running buildContextOverflowRecoveryText with patched code:

=== 1M context (openrouter/qwen3.6-plus) ===
⚠️ Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting `agents.defaults.compaction.reserveTokensFloor` to 100000 or higher in your config.

=== 200k context (claude-sonnet-4-6) ===
⚠️ Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting `agents.defaults.compaction.reserveTokensFloor` to 50000 or higher in your config.

=== 100k context (gpt-4o) ===
⚠️ Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting `agents.defaults.compaction.reserveTokensFloor` to 35000 or higher in your config.

=== 32k context (ollama/qwen3.5) ===
⚠️ Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting `agents.defaults.compaction.reserveTokensFloor` to 20000 or higher in your config.

=== undefined (no model metadata) ===
⚠️ Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting `agents.defaults.compaction.reserveTokensFloor` to 20000 or higher in your config.

Observed result after fix: The recovery hint now recommends context-appropriate values: 100000 for 1M, 50000 for 200k, 35000 for 100k, and the default 20000 for small/undefined context windows. Before this patch, all cases showed the hardcoded 20000.

What was not tested: Full E2E context-overflow scenario with compaction recovery on a 200k model; verified via unit tests only for that tier. Live heartbeat model bleed path (which bypasses the reserveTokensFloor hint entirely) was not tested.

Fixes #65839

clawsweeper · 2026-05-20T02:25:42Z

Codex review: needs maintainer review before merge. Reviewed May 25, 2026, 7:16 PM ET / 23:16 UTC.

Summary
The PR changes context-overflow recovery text to recommend reserveTokensFloor values from the effective model context window instead of always suggesting 20000.

PR surface: Source +106, Tests +507. Total +613 across 2 files.

Reproducibility: yes. at source level: current main and v2026.5.22 both hardcode the reserveTokensFloor hint to 20000, matching the linked bug. I did not run the app or tests in this read-only review.

Review metrics: none identified.

Merge readiness
Overall: 🦞 diamond lobster
Proof: 🦞 diamond lobster
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Risk before merge

I did not execute the focused Vitest command in this read-only review; merge should still wait for required checks on ea02a9602e4503e8cc36afdda82e81eb504609fa.

Maintainer options:

Decide the mitigation before merge
Land this focused PR after required CI is green so the linked issue can close through the PR's closing reference.
Pause or close
Do not merge this PR until maintainers decide whether the risk is worth taking.

Next step before merge
No ClawSweeper repair lane is needed; the remaining action is maintainer review and normal CI gating for the current head.

Security
Cleared: The diff only touches the auto-reply recovery helper and colocated tests; it does not add dependencies, workflows, scripts, permissions, or secret-handling changes.

Review details

Best possible solution:

Land this focused PR after required CI is green so the linked issue can close through the PR's closing reference.

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

Yes at source level: current main and v2026.5.22 both hardcode the reserveTokensFloor hint to 20000, matching the linked bug. I did not run the app or tests in this read-only review.

Is this the best way to solve the issue?

Yes; the PR keeps the existing config key and derives the hint from the effective runtime context window, including existing agent context caps and session/runtime fallback behavior, rather than adding a new config surface.

AGENTS.md: found and applied where relevant.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 65a210553b6c.

Label changes

Label changes:

add rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🦞 diamond lobster and patch quality is 🦞 diamond lobster.
remove rating: 🐚 platinum hermit: Current PR rating is rating: 🦞 diamond lobster, so this older rating label is no longer current.

Label justifications:

P2: This fixes a normal-priority user-facing recovery hint for context overflow without changing persisted defaults or core routing behavior.
rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🦞 diamond lobster and patch quality is 🦞 diamond lobster.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (terminal): The PR body includes after-fix terminal output from the patched recovery text helper showing the expected tiered hints for 1M, 200k, 100k, 32k, and unknown context windows.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes after-fix terminal output from the patched recovery text helper showing the expected tiered hints for 1M, 200k, 100k, 32k, and unknown context windows.

Evidence reviewed

PR surface:

Source +106, Tests +507. Total +613 across 2 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	119	13	+106
Tests	1	507	0	+507
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	2	626	13	+613

What I checked:

Repository policy read: The full root policy was read and applied, including PR review, compatibility, proof, and source-history requirements. (AGENTS.md:1)
Scoped policy check: No scoped AGENTS.md exists under src/auto-reply, and the only maintainer note found is Telegram-specific, so root policy owns this review.
Current main still has the reported behavior: At the provided current main SHA, the overflow reset hint still hardcodes reserveTokensFloor to 20000. (src/auto-reply/reply/agent-runner-execution.ts:717, 65a210553b6c)
Latest release still has the reported behavior: The latest release tag also hardcodes the same 20000 hint, so the PR is not obsolete from shipped behavior. (src/auto-reply/reply/agent-runner-execution.ts:716, a374c3a5bfd5)
PR implements dynamic reserve floor tiers: The PR head adds computeContextAwareReserveTokensFloor and builds the reset hint with tiered values for 1M, 200k, 100k, and smaller or unknown context windows. (src/auto-reply/reply/agent-runner-execution.ts:717, ea02a9602e45)
PR threads effective runtime model into overflow hints: The PR passes the attempted runtime provider/model into both embedded overflow and compaction-failure recovery paths, so fallback candidates can drive the hint. (src/auto-reply/reply/agent-runner-execution.ts:2374, ea02a9602e45)

Likely related people:

fuller-stack-dev: Commit cd1cae5be9ba76763e696784ff2effa534e00052 added preserved-session overflow recovery guidance and tests around the same buildContextOverflowRecoveryText path. (role: introduced adjacent recovery behavior; confidence: medium; commits: cd1cae5be9ba; files: src/auto-reply/reply/agent-runner-execution.ts, src/auto-reply/reply/agent-runner-execution.test.ts)
vincentkoc: Commit 04061bc801891060fb34a425af5fc08c38dd1ad3 changed the heartbeat context-window fallback logic that shares the same overflow hint decision surface. (role: recent area contributor; confidence: high; commits: 04061bc80189; files: src/auto-reply/reply/agent-runner-execution.ts, src/auto-reply/reply/agent-runner-execution.test.ts)
steipete: The current PR head was force-pushed by steipete and includes maintainer follow-up commits tightening runtime candidate, stale-session, and heartbeat-hint handling. (role: recent PR follow-up owner; confidence: high; commits: b3203d4fd0b4, 1ec913025cc9, 9d28cba8cea4; files: src/auto-reply/reply/agent-runner-execution.ts, src/auto-reply/reply/agent-runner-execution.test.ts)

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

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.

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.

clawsweeper · 2026-05-20T02:41:09Z

ClawSweeper PR egg

✨ Hatched: 🥚 common Neon Crabkin

Hatch command

Comment @clawsweeper hatch when this PR is hatchable.

Hatchability rules:

Merged PRs are hatchable.
Open PRs are hatchable when they are status: 👀 ready for maintainer look, status: 🚀 automerge armed, or labeled clawsweeper:automerge.
Closed unmerged PRs are hatchable only when one of those hatchable labels is still present in the durable record.

Rarity: 🥚 common.
Trait: sparkles near resolved comments.
Image traits: location review cove; accessory proof snapshot camera; palette moss green and polished brass; mood mischievous; pose nestled inside a glowing shell; shell matte ceramic shell; lighting gentle morning glow; background gentle dashboard dots.
Share on X: post this hatch
Copy: My PR egg hatched a 🥚 common Neon Crabkin in ClawSweeper.

What is this egg doing here?

Eggs appear after the PR passes real-behavior proof. It is here for vibes, not verdicts: it does not change labels, ratings, merge decisions, or automation.
The shell reacts to review momentum: open follow-up work warms it up, re-review makes it wobble, and a clean final review lets it hatch.
Hatchability usually comes from sufficient real-behavior proof, no blocking P0/P1/P2 findings, no security attention needed, and clean correctness. A merged PR is already final, so merge makes the egg hatchable independently.
The hatch is seeded from this repository and PR number, so the same PR keeps the same creature; the reviewed head SHA can only change safe visual details.
Rarity is just collectible sparkle: 🥚 common, 🌱 uncommon, 💎 rare, ✨ glimmer, and 🌈 legendary.

clawsweeper · 2026-05-20T03:42:08Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

State: Complete
Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
Run: https://github.com/openclaw/clawsweeper/actions/runs/26139864704
Updated: 2026-05-20T03:50:49.014Z

clawsweeper · 2026-05-20T04:14:07Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

State: Complete
Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
Run: https://github.com/openclaw/clawsweeper/actions/runs/26140861010
Updated: 2026-05-20T04:22:09.736Z

Replace hardcoded 20000 reserveTokensFloor suggestion with dynamic tier-based values based on the model context window size: - 100000 for context windows >= 1M tokens - 50000 for context windows >= 200k tokens - 35000 for context windows >= 100k tokens - 20000 for smaller/undefined context windows Uses session contextTokens as fallback when model metadata is unavailable, with model metadata taking precedence over stale session entries. Fixes openclaw#65839

When agents.defaults.contextTokens or per-agent contextTokens caps the effective context window below the model's full window, the hint should reflect the capped value, not the raw model window. Uses Math.min(agentCap, modelWindow) to compute the effective context, matching resolveContextTokens() in model-selection.ts.

When the primary model's context window cannot be resolved from model metadata, fall back to the session entry's modelProvider/model to look up context window before using session contextTokens numeric.

When a large primary model falls back to a smaller runtime model that overflows, the recovery hint should recommend a reserveTokensFloor based on the actual runtime model's context window, not the primary model's larger window. Thread runtimeProvider/runtimeModel into buildContextOverflowRecoveryText and resolveContextWindowForCompactionHint, with priority: 1. Runtime/fallback model (the one that actually overflowed) 2. Session entry model (when runtime not provided) 3. Primary model (fallback when runtime unknown) 4. Agent contextTokens cap applied via Math.min 5. Session contextTokens numeric value

…4399) Use the effective runtime/model context when computing overflow recovery reserveTokensFloor hints, including uncataloged runtime refs, stale session windows, and heartbeat fallback cases. Verification: - pnpm test src/auto-reply/reply/agent-runner-execution.test.ts - autoreview clean on final focused fixup; prior accepted findings addressed before push. - CI passed on head e25b3e8 after rerunning cancelled jobs: preflight, critical quality network-runtime-boundary, security high, checks, Real behavior proof. Co-authored-by: tanshanshan <tanshanshan@users.noreply.github.com>

…to PR-head wholesale Per pr-review-toolkit:code-reviewer finding: my prior shape-only port left computeContextAwareReserveTokensFloor as dead code and runtimeProvider/ runtimeModel params unused, regressing upstream openclaw#84399 ("use context-aware overflow reserve hints"). The 10 test failures attributed in RESOLUTION-TRAIL §11 to "upstream-class loss" included this real regression of an upstream feature, not just an upstream-only test that PR-head intentionally lacked. Full port-back applied to PR-head's wholesale-taken src/auto-reply/reply/agent-runner-execution.ts: - resolveContextWindowForCompactionHint() helper added (resolves context window from runtime / session / primary / agent-cap, in that priority order) - buildContextOverflowResetHint() helper added (dynamic per-window reset hint using computeContextAwareReserveTokensFloor) - buildContextOverflowRecoveryText() body rewritten to compute the primary context window and select heartbeat-bleed-hint vs reset-hint per upstream - Static CONTEXT_OVERFLOW_RESET_HINT constant removed (was unused after rewrite) - CLI runner runParams now carries suppressNextUserMessagePersistence, onUserMessagePersisted, and userTurnTranscriptRecorder (upstream's prepared-CLI-user-turn boundary) - attemptedRuntimeProvider / attemptedRuntimeModel tracking added at function scope; updated inside the run() callback per candidate attempt - Three buildContextOverflowRecoveryText callsites now thread runtimeProvider: attemptedRuntimeProvider / runtimeModel: attemptedRuntimeModel (was fallbackProvider/fallbackModel, which only updates on success) - Added unconditional `if (isCompactionFailure)` fallback branch with preserveSessionMapping: true mirroring upstream, AFTER PR-head's existing resetSessionAfterCompactionFailure-gated branch. Preserves PR-head's reset- attempt feature when the callback returns true; falls back to upstream's always-show-recovery-text behavior when reset returns false. Test gate: src/auto-reply/reply/agent-runner-execution.test.ts now 149/149 PASS (was 10 failing). Tsgo / umbrella check / build all green. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…fixture corrections Supersedes prior fold-3-defer-2 ambiguity. Per figs verbatim canon at PR #792 issuecomment-4551691544 (Elliott byte-walk correction): "we ship feature complete. thats a weird deferral" — fold ALL 5 #793 substantive deltas, no active-vs-dormant split. Plus paired test corrections that BOTH lanes shipped (per Elliott byte-walk): 1. **`src/agents/pi-embedded-runner/run/failover-policy.ts`** — adopt #793's `harnessOwnedTimeout` extraction + early-return `continue_normal` shape. Reverts my earlier `bf21bbc4b8`-shape revert. Cohort lock: full fold. 2. **`src/infra/session-cost-usage.ts`** — adopt #793's `includeCheckpoints` typed flag at helper source. Reverts my earlier `bf21bbc4b8`-shape revert (targeted skip in `loadCostUsageSummary` loop). Cleaner abstraction. 3. **`src/agents/pi-tools.workspace-paths.test.ts`** — fixture rename `memory/2026-05-20.md` → `memory/new-note.md` to de-stamp date-baked fixture. Test assertion shape unchanged (`fs.readFile(...).resolves.toBe(...)`). Both lanes converged on this rename per Elliott `4551691544`. NOT a mask. 4. **`src/agents/pi-embedded-runner/run.cross-provider-fallback-error-context.test.ts`** — paired test update matching merged production behavior under openclaw#84399 + #793 harnessOwnsTransport shape. Assertion now asserts what production actually does (`.resolves.toMatchObject({payloads: [{isError: true, text: stringContaining("Request timed out")}]})`). Both lanes shipped paired test updates for their contract changes per Elliott byte-walk. NOT a mask. Cael's earlier "Ronan masked tests / Cael honest-fails" framing was wrong; Elliott byte-walked and self-corrected (PR comments 4551691544 + 4551691738). This commit aligns cael's test surface to the cohort convergence. Test gate: 38/38 in the 2 previously-red files; tsgo green. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…4399) Use the effective runtime/model context when computing overflow recovery reserveTokensFloor hints, including uncataloged runtime refs, stale session windows, and heartbeat fallback cases. Verification: - pnpm test src/auto-reply/reply/agent-runner-execution.test.ts - autoreview clean on final focused fixup; prior accepted findings addressed before push. - CI passed on head e25b3e8 after rerunning cancelled jobs: preflight, critical quality network-runtime-boundary, security high, checks, Real behavior proof. Co-authored-by: tanshanshan <tanshanshan@users.noreply.github.com>

clawsweeper · 2026-06-07T19:41:30Z

🦞👀
ClawSweeper picked this up.

Command router queued. I will update this comment with the next step.

…4399) Use the effective runtime/model context when computing overflow recovery reserveTokensFloor hints, including uncataloged runtime refs, stale session windows, and heartbeat fallback cases. Verification: - pnpm test src/auto-reply/reply/agent-runner-execution.test.ts - autoreview clean on final focused fixup; prior accepted findings addressed before push. - CI passed on head e25b3e8 after rerunning cancelled jobs: preflight, critical quality network-runtime-boundary, security high, checks, Real behavior proof. Co-authored-by: tanshanshan <tanshanshan@users.noreply.github.com>

openclaw-barnacle Bot added size: M triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 20, 2026

openclaw-barnacle Bot added proof: supplied External PR includes structured after-fix real behavior proof. and removed triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels May 20, 2026

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 20, 2026

clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 20, 2026

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 20, 2026

clawsweeper Bot added the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 20, 2026

This comment was marked as low quality.

Sign in to view

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 20, 2026

clawsweeper Bot mentioned this pull request May 25, 2026

fix: use context-aware reserveTokensFloor in overflow recovery hint #86401

Closed

1 task

steipete force-pushed the fix/context-aware-reserve-tokens-floor-v2 branch from e76b8cb to ea02a96 Compare May 25, 2026 23:09

openclaw-barnacle Bot added size: L and removed size: M labels May 25, 2026

clawsweeper Bot added rating: 🦞 diamond lobster Very strong PR readiness with only minor maintainer review expected. and removed rating: 🐚 platinum hermit Good normal PR readiness with ordinary maintainer review expected. labels May 25, 2026

tanshanshan and others added 9 commits May 26, 2026 00:21

fix: use fallback model from session when primary model is unresolved

4330aab

When the primary model's context window cannot be resolved from model metadata, fall back to the session entry's modelProvider/model to look up context window before using session contextTokens numeric.

fix(auto-reply): reuse context cap helper for overflow hints

e8d89c1

fix(auto-reply): track overflow hint runtime candidates

86f26e8

fix(auto-reply): avoid primary overflow hint fallback for runtime refs

24005fc

fix(auto-reply): ignore stale session windows for runtime hints

809f8dd

fix(auto-reply): guard heartbeat hints for runtime refs

e25b3e8

steipete force-pushed the fix/context-aware-reserve-tokens-floor-v2 branch from ea02a96 to e25b3e8 Compare May 25, 2026 23:23

openclaw-barnacle Bot removed the proof: sufficient ClawSweeper judged the real behavior proof convincing. label May 25, 2026

steipete merged commit 3a4f2b1 into openclaw:main May 25, 2026
109 of 162 checks passed

github-actions Bot mentioned this pull request May 26, 2026

📡 Upstream Digest — 2026-05-26 02:30 UTC curtismercier/openclaw-mods#948

Open

This was referenced May 27, 2026

[wo-pr85651-driftcure] ronan-copilot lane journal karmaterminal/openclaw#789

Open

[wo-pr85651-driftcure] cael-claude lane journal karmaterminal/openclaw#790

Open

This was referenced May 27, 2026

DO NOT MERGE — draft: PR#85651 drift-cure candidate (cael-seat claude lane) #790 karmaterminal/openclaw#792

Draft

DO NOT MERGE — draft: PR#85651 drift-cure candidate (ronan-seat copilot lane) #789 karmaterminal/openclaw#793

Draft

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: use context-aware reserveTokensFloor in overflow recovery hint#84399

fix: use context-aware reserveTokensFloor in overflow recovery hint#84399
steipete merged 9 commits into
openclaw:mainfrom
tanshanshan:fix/context-aware-reserve-tokens-floor-v2

tanshanshan commented May 20, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 20, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented May 20, 2026 •

edited

Loading

Uh oh!

This comment was marked as low quality.

clawsweeper Bot commented May 20, 2026 •

edited

Loading

Uh oh!

This comment was marked as low quality.

clawsweeper Bot commented May 20, 2026 •

edited

Loading

Uh oh!

Uh oh!

clawsweeper Bot commented Jun 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

tanshanshan commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Problem

Solution

Changes

Verification

Real behavior proof

Uh oh!

clawsweeper Bot commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

clawsweeper Bot commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Hatch command

Uh oh!

This comment was marked as low quality.

clawsweeper Bot commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment was marked as low quality.

clawsweeper Bot commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

clawsweeper Bot commented Jun 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

tanshanshan commented May 20, 2026 •

edited

Loading

clawsweeper Bot commented May 20, 2026 •

edited

Loading

clawsweeper Bot commented May 20, 2026 •

edited

Loading

clawsweeper Bot commented May 20, 2026 •

edited

Loading

clawsweeper Bot commented May 20, 2026 •

edited

Loading