fix(status): comprehensive context window resolution — fixes slash-model 200K bug (closes #39857, #92760) by samson910022 · Pull Request #92772 · openclaw/openclaw

samson910022 · 2026-06-13T18:37:28Z

Description

Comprehensive fix for /status showing ?/200k (DEFAULT_CONTEXT_TOKENS) for proxy/aggregator models with slash-containing IDs (e.g., opencode-go/deepseek-v4-flash, openrouter/anthropic/claude-sonnet-4).

Diagnosis: 7-layer fix stack �X 6 code patches in context.ts/status-message.ts/status.summary.runtime.ts + 1 config gap.

Root Cause

The discovery cache (
untime.json registry) was never populated during normal gateway operation. All callers pass cfg (truthy), setting skipRuntimeConfigLoad = true, which bypasses ensureContextWindowCacheLoaded(). The cache only contained bare-key entries from primeConfiguredContextWindows(), causing cache miss �� DEFAULT_CONTEXT_TOKENS (200K).

Even if the cache had been populated:

The !ref.model.includes("/") guard blocked qualified key lookup for slash-containing model IDs
status-message.ts did not extract embedded provider from slash-model IDs
The CLI status.summary.runtime.ts had its own standalone copy of the resolution function

Fixes (11 commits on ix/comprehensive-context-window-resolution)

Commit	Fix	File
`980f60a`	PR #92424: Fix stale session modelProvider after /model switch	agent-session.ts
`c89b83d`	PR #92709: Remove !ref.model.includes("/") guard blocking qualified key lookup	context.ts
`d6be5de`	Fix A: Extract embedded provider from slash model ID in legacy sessions	status-message.ts
`60b5bc2`	Fix B: Store provider-qualified keys (provider/model) in discovery cache	context.ts
`5295b09`	Fix D: Kick off background discovery on first model ref resolution	context.ts
`9736c88`	Move discovery kickoff inside if(ref) guard per code review	context.ts
`58270ec`	IMP-1: Move kickoff to TOP of if(ref) �X fires for ALL resolver paths	context.ts
`44ffb37`	CRIT-1/2: CLI status delegation �X remove standalone copy, import from context module	status.summary.runtime.ts

Verification

Subagent audit (source ? dist alignment)

All 7 patches confirmed present in both ~/projects/openclaw/src/ and installed dist (context-ClaruzGd.js).

Manual test (user, 2026-06-14 02:35)

\
?? Session: opencode/nemotron-3-ultra-free
?? Context: 401/1.0m (0%) �� Correct 1.0M for this model

?? Session: opencode-go/deepseek-v4-flash
?? Context: 0/1.0m (0%) �� WAS 200K, NOW 1.0M ? FIXED

?? Session: opencode/deepseek-v4-flash-free
?? Context: 611/200k (0%) �� Correct 200K for free tier
\\

TUI verification

TUI correctly shows live context tracking (no ? placeholder) and correct window for each model.

Related Issues

Closes Status/session context window can over-report the selected model's actual window #39857 �X Context window shows 200K for proxy models with slash
Closes CLI openclaw status shows 200K context window due to standalone resolution copy in status.summary.runtime.ts #92760 �X CLI openclaw status uses standalone copy of resolution function
Related fix: cap session context overrides by model window #90889 �X Config context window override (complementary, non-overlapping)
Related Session-level AgentSession.this.model snapshot is never refreshed after /model switch (affects contextWindow, reasoning, thinkingLevelMap, branch summary) #92415 �X Session model snapshot stale after /model switch (fixed by fix(agents): resolve fresh model from registry for post-turn reads after /model switch (fixes #92415) #92424)

…ter /model switch All 8 post-turn reads of this.model in AgentSession (checkCompaction, isRetryableError, getContextUsage, getAvailableThinkingLevels, supportsThinking, clampThinkingLevel, compact, generateBranchSummary) now resolve the model fresh from sessionModelRegistry instead of reading the stale agent.state.model snapshot. This ensures contextWindow, reasoning, and thinkingLevelMap reflect the current model after a /model switch. Fixes openclaw#92415

…test and comment

…ts in resolve-fresh-model test

- Remove dead code: use resolveFreshModel()! in getAvailableThinkingLevels since the early return guarantees this.model is truthy - Add JSDoc explaining why resolveFreshModel() does not use refreshCurrentModelFromRegistry() (mutation vs transient read) - Add defensive test for registry returning different provider/id - Add integration test verifying compaction uses fresh contextWindow

… OpenRouter proxy models) Remove the `!ref.model.includes("/")` guard in `resolveContextTokensForModel()` which blocked qualified cache key lookup for models like `anthropic/claude-sonnet-4` accessed through proxy providers. Two call sites fixed: 1. Explicit provider path (params.provider && ref) -- safe because provider is explicitly given 2. Last resort implicit provider path (!params.provider && ref) -- bare key already failed, trying qualified key has no downside Fixes: max context showing DEFAULT_CONTEXT_TOKENS (200K) instead of actual model's context window for proxy models

…t window resolution When a session entry has model="openrouter/anthropic/claude-sonnet-4" but entry.modelProvider is empty, buildStatusMessage() previously set contextLookupProvider = undefined, causing resolveContextTokensForModel() to skip the provider-qualified cache key lookup. Now extract the embedded provider (first slash segment) and pass it to the resolution chain so the correct context window is found. Combined with PR openclaw#92424's resolveFreshModel() fix and PR openclaw#92709's slash-guard removal in resolveContextTokensForModel(), this gives comprehensive coverage for proxy model context window resolution. Refs: openclaw#39857, openclaw#90889, openclaw#92424, openclaw#92709

…d-model cache

…ensForModel to populate provider-qualified cache keys

…real context.ts function

… cover all resolver paths

clawsweeper · 2026-06-13T18:39:38Z

Codex review: needs real behavior proof before merge. Reviewed June 13, 2026, 2:44 PM ET / 18:44 UTC.

Summary
This PR combines AgentSession fresh-model reads, shared context-window cache/provider-qualified lookup changes, CLI status resolver delegation, and status-message slash-model provider extraction for context-window reporting.

PR surface: Source -16, Tests +118. Total +102 across 5 files.

Reproducibility: yes. Source and adjacent tests show the patch would change existing model-only slash-ID status behavior, and current resolver code documents that allowAsyncLoad:false paths must not start background discovery or models.json writes.

Review metrics: 1 noteworthy metric.

Read-only resolver side effect: 1 unguarded warmup call added. This matters because current resolver code explicitly separates allowAsyncLoad:false read-only lookups from background discovery and models.json writes.

Stored data model
Persistent data-model change detected: persistent cache schema: src/agents/context.ts, serialized state: src/agents/sessions/agent-session.resolve-fresh-model.test.ts, unknown-data-model-change: src/agents/sessions/agent-session.resolve-fresh-model.test.ts, unknown-data-model-change: src/agents/sessions/agent-session.ts. Confirm migration or upgrade compatibility proof before merge.

Merge readiness
Overall: 🧂 unranked krab
Proof: 🦪 silver shellfish
Patch quality: 🧂 unranked krab
Result: blocked until stronger real behavior proof is added.

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

Rank-up moves:

[P1] Fix the two blocking resolver/status findings and keep or add regression tests for both invariants.
[P1] Add exact after-fix real behavior proof with redacted terminal output or screenshots showing the command, model setup, and observed status result.

Proof guidance:

[P1] Needs stronger real behavior proof before merge: The body includes copied after-fix status output, but it lacks exact commands/environment or an inspectable artifact and does not cover the side-effect-sensitive read-only paths; add redacted terminal output or screenshots and update the PR body to trigger re-review.

Risk before merge

[P1] Merging as-is can misreport OpenRouter-style slash model IDs by treating the first model segment as an OpenClaw provider and applying another provider's configured context window.
[P1] Merging as-is can make read-only status/summary paths start background model discovery, credential-backed registry loading, and models.json/plugin-catalog writes.
[P1] The body includes copied after-fix output, but the contributor still needs exact real behavior proof with commands, environment, and redacted terminal output or screenshots.

Maintainer options:

Fix the resolver and slash-id invariants first (recommended)
Keep model-only slash IDs raw unless selected/fallback state proves provider ownership, and preserve allowAsyncLoad:false as a no-discovery/no-models-json-write path with regression coverage.
Accept status-side side effects deliberately
Maintainers could explicitly decide that /status should warm model discovery, but the PR should then document and test the new operator-visible models.json/auth-registry behavior.
Split the combined branch
If the combined stack is too risky, keep the AgentSession and CLI delegation pieces as narrower PRs and leave slash-model cache discovery for a focused follow-up.

Copy recommended automerge instruction

@clawsweeper automerge

Special instructions:
Preserve model-only slash IDs in status-message unless fallback/selected model state proves provider ownership, move context-window discovery kickoff behind allowAsyncLoad true or a lifecycle-owned warmup, and add regression tests for OpenRouter-style slash IDs plus read-only no models.json side effects.

Next step before merge

[P1] The remaining blockers are concrete code/test repairs in the touched files and do not require a product decision.

Security
Cleared: No new dependencies, workflows, package metadata, or secret-output paths were introduced; the credential/discovery side effect is handled as a functional merge risk.

Review findings

[P1] Preserve model-only slash IDs in status — src/status/status-message.ts:628-629
[P1] Honor read-only context lookups — src/agents/context.ts:397

Review details

Best possible solution:

Repair the branch so status keeps raw model-only slash IDs unless provider ownership is proven, and move context-window cache warmup to an allowAsyncLoad/lifecycle-owned path instead of read-only resolver calls.

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

Yes. Source and adjacent tests show the patch would change existing model-only slash-ID status behavior, and current resolver code documents that allowAsyncLoad:false paths must not start background discovery or models.json writes.

Is this the best way to solve the issue?

No. Delegating CLI status to the shared resolver is a good direction, but the PR is not the best fix until it preserves the slash-ID ownership invariant and keeps cache discovery out of read-only resolver calls.

Full review comments:

[P1] Preserve model-only slash IDs in status — src/status/status-message.ts:628-629
This branch changes the missing-modelProvider slash case to pass provider as the first segment and model as the remainder. Current status code and tests deliberately keep those entries as model-only when the slash can be part of an upstream model ID; for google/gemini-2.5-pro with no modelProvider, the new path scans models.providers.google and reports Google's configured window instead of the raw OpenRouter-style cache entry. Keep the existing selected/fallback proof before splitting the model string.
Confidence: 0.93
[P1] Honor read-only context lookups — src/agents/context.ts:397
resolveContextTokensForModel now kicks off ensureContextWindowCacheLoaded() whenever ref exists, including callers that pass cfg and allowAsyncLoad: false. Current lookupContextTokens documents that those read-only callers must not start background model discovery or models.json writes, and ensureContextWindowCacheLoaded can call ensureOpenClawModelsJson plus credential-backed model discovery. Gate this behind allowAsyncLoad or a lifecycle-owned warmup path.
Confidence: 0.91

Overall correctness: patch is incorrect
Overall confidence: 0.9

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 13a079b3f846.

Label changes

Label changes:

add P2: This is a normal-priority status/context-window correctness PR with limited blast radius but real user-facing diagnostics impact.
add merge-risk: 🚨 compatibility: The PR changes established slash-model lookup semantics and read-only status side effects that existing users may rely on.
add merge-risk: 🚨 session-state: Incorrect context-window resolution can affect displayed and persisted session context accounting.
add merge-risk: 🚨 auth-provider: The PR infers provider identity from slash-containing model IDs and can select the wrong provider context/auth-discovery path.
add rating: 🧂 unranked krab: Overall readiness is 🧂 unranked krab; proof is 🦪 silver shellfish and patch quality is 🧂 unranked krab.
add status: 📣 needs proof: The PR needs real behavior proof before ClawSweeper can clear the contributor ask. Needs stronger real behavior proof before merge: The body includes copied after-fix status output, but it lacks exact commands/environment or an inspectable artifact and does not cover the side-effect-sensitive read-only paths; add redacted terminal output or screenshots and update the PR body to trigger re-review.

Label justifications:

P2: This is a normal-priority status/context-window correctness PR with limited blast radius but real user-facing diagnostics impact.
merge-risk: 🚨 compatibility: The PR changes established slash-model lookup semantics and read-only status side effects that existing users may rely on.
merge-risk: 🚨 session-state: Incorrect context-window resolution can affect displayed and persisted session context accounting.
merge-risk: 🚨 auth-provider: The PR infers provider identity from slash-containing model IDs and can select the wrong provider context/auth-discovery path.
rating: 🧂 unranked krab: Overall readiness is 🧂 unranked krab; proof is 🦪 silver shellfish and patch quality is 🧂 unranked krab.
status: 📣 needs proof: The PR needs real behavior proof before ClawSweeper can clear the contributor ask. Needs stronger real behavior proof before merge: The body includes copied after-fix status output, but it lacks exact commands/environment or an inspectable artifact and does not cover the side-effect-sensitive read-only paths; add redacted terminal output or screenshots and update the PR body to trigger re-review.

Evidence reviewed

PR surface:

Source -16, Tests +118. Total +102 across 5 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	4	63	79	-16
Tests	1	118	0	+118
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	5	181	79	+102

Acceptance criteria:

[P1] node scripts/run-vitest.mjs src/auto-reply/status.test.ts src/agents/context.lookup.test.ts src/commands/status.summary.runtime.test.ts.
[P1] node scripts/run-vitest.mjs src/agents/context.test.ts src/agents/sessions/agent-session.resolve-fresh-model.test.ts.

What I checked:

PR changes model-provider splitting in status rendering: The PR changes the missing-modelProvider slash-model branch to split the first segment into contextLookupProvider and pass the remainder as contextLookupModel, which changes current model-only lookup behavior for upstream model IDs that contain slashes. (src/status/status-message.ts:628, 58270ec96e98)
Current main preserves raw model-only slash IDs: Current status-message code keeps raw model-only lookup for OpenRouter-style slash IDs unless the slash ID is proven to be the selected model or active fallback target. (src/status/status-message.ts:612, 13a079b3f846)
Adjacent status test encodes the invariant: Existing status coverage expects a missing-modelProvider runtime model like google/gemini-2.5-pro to use the raw model-only cache entry and not Google's configured provider window. (src/auto-reply/status.test.ts:2220, 13a079b3f846)
PR adds unguarded background discovery in the resolver: The PR calls ensureContextWindowCacheLoaded whenever a model ref is resolvable, before checking allowAsyncLoad or read-only status-call semantics. (src/agents/context.ts:397, 58270ec96e98)
Current resolver documents the read-only contract: Current lookupContextTokens explicitly says allowAsyncLoad:false callers should not start background model discovery or models.json writes. (src/agents/context.ts:217, 13a079b3f846)
Background discovery can write models.json: ensureContextWindowCacheLoaded reaches ensureOpenClawModelsJson, whose implementation may create the agent directory and write models.json/plugin catalog sidecars. (src/agents/models-config.ts:445, 13a079b3f846)

Likely related people:

steipete: Recent GitHub commit history shows repeated work on context-window, status-message, and model-runtime behavior in the same surfaces. (role: recent area contributor; confidence: high; commits: 0e7b5c34292c, 9ead0ae9219e, edb920b857ce; files: src/agents/context.ts, src/status/status-message.ts)
RomneyDa: Authored the context-cache warmup removal that stopped ensureOpenClawModelsJson/provider discovery from running on TUI/status-style imports. (role: perf/runtime side-effect contributor; confidence: high; commits: b79effefee92, d91ef6bb171d; files: src/agents/context.ts, src/agents/context.lookup.test.ts)
zhangguiping-xydt: The checked-out shallow history blames the current context/status/status-summary implementations to the latest main commit touching agent runtime prompt/session identity surfaces. (role: current-main adjacent contributor; confidence: medium; commits: 78c66742ab97; files: src/agents/context.ts, src/status/status-message.ts, src/commands/status.summary.runtime.ts)
giodl73-repo: Recent status-message history shows this author added estimated context budget status behavior adjacent to the status context display path. (role: status context-budget contributor; confidence: medium; commits: 05c6e7a55391; files: src/status/status-message.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.

samson910022 · 2026-06-13T18:52:36Z

Superseded — committing a narrower branch focused on status display only.

…/status on first call The first /status after gateway restart shows DEFAULT_CONTEXT_TOKENS (200K) for ALL providers because async model discovery hasn't completed yet. resolveStatusRuntimeContextTokens already has a lazy guard (await ensureContextWindowCacheLoaded()) but the first call still pays cold-start latency on the user-facing /status path. Fix: fire the warmup eagerly during gateway startup inside startGatewayEarlyRuntime(), before the HTTP server starts listening. The call uses void semantics (fire-and-forget) to avoid blocking the startup sequence. The function is already idempotent via a singleton loadPromise guard and catches all errors internally. This is a companion optimization to the lazy-guard already present in resolveStatusRuntimeContextTokens (src/status/status-text.ts). Together they ensure: - First /status: cache already warm from startup - Subsequent /status: shares resolved singleton promise (microtask) - Mid-session /model switch: lazy guard triggers re-check via config primeConfiguredContextWindows with backoff - Concurrent /status calls: all share one discovery pass Relates to: openclaw#39857, openclaw#92760 Companion to: openclaw#92775 (CLI fix) Supersedes: openclaw#92772 (comprehensive branch)

liuhao1024 and others added 11 commits June 13, 2026 23:38

fix: use thinkingBudgets instead of non-existent thinkingLevelMap in …

980f60a

…test and comment

fix(test): use thinkingLevelMap instead of non-existent thinkingBudge…

fb1d56a

…ts in resolve-fresh-model test

fix(agents): store provider-qualified context-token keys in discovere…

60b5bc2

…d-model cache

fix(agents): kick off background model discovery in resolveContextTok…

5295b09

…ensForModel to populate provider-qualified cache keys

refactor: move async discovery kickoff inside ref guard per code review

9736c88

fix(commands): delegate status.summary.runtime context resolution to …

44ffb37

…real context.ts function

fix(agents): move ensureContextWindowCacheLoaded to top of if(ref) to…

58270ec

… cover all resolver paths

openclaw-barnacle Bot added commands Command implementations agents Agent runtime and tooling size: M labels Jun 13, 2026

openclaw-barnacle Bot added the triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. label Jun 13, 2026

clawsweeper Bot added rating: 🧂 unranked krab Not merge-ready due to missing proof or serious correctness/safety concerns. status: 📣 needs proof The PR needs real behavior proof before ClawSweeper can clear the contributor ask. labels Jun 13, 2026

samson910022 closed this Jun 13, 2026

samson910022 mentioned this pull request Jun 13, 2026

fix(status): warm context window cache at startup for correct /status on first call #92775

Closed

samson910022 mentioned this pull request Jun 14, 2026

fix(status): warm model context-window cache before /status reads it #92905

Closed

samson910022 deleted the fix/comprehensive-context-window-resolution branch June 16, 2026 16:07

clawsweeper Bot mentioned this pull request Jun 18, 2026

fix(cli): resolve 200K context window fallback in status command #94210

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(status): comprehensive context window resolution — fixes slash-model 200K bug (closes #39857, #92760)#92772

fix(status): comprehensive context window resolution — fixes slash-model 200K bug (closes #39857, #92760)#92772
samson910022 wants to merge 11 commits into
openclaw:mainfrom
samson910022:fix/comprehensive-context-window-resolution

samson910022 commented Jun 13, 2026

Uh oh!

clawsweeper Bot commented Jun 13, 2026 •

edited

Loading

Uh oh!

samson910022 commented Jun 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Uh oh!

Conversation

samson910022 commented Jun 13, 2026

Description

Root Cause

Fixes (11 commits on ix/comprehensive-context-window-resolution)

Verification

Subagent audit (source ? dist alignment)

Manual test (user, 2026-06-14 02:35)

TUI verification

Related Issues

Uh oh!

clawsweeper Bot commented Jun 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

samson910022 commented Jun 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

clawsweeper Bot commented Jun 13, 2026 •

edited

Loading