fix(sessions): provider-qualified context limits (#62472) by neeravmakwana · Pull Request #62493 · openclaw/openclaw

neeravmakwana · 2026-04-07T12:35:32Z

Summary

Problem: Session usage persistence, inline directive handling, and memory-flush / preflight compaction sizing used bare model ids (lookupContextTokens / lookupCachedContextTokens) for context limits. When several providers expose the same model id with different configured context windows, the wrong limit could be written to sessionEntry.contextTokens and affect /status, compaction, and flush behavior ([Bug]: Context token / context window can be persisted incorrectly when multiple providers share the same model id, causing wrong /status values and potentially premature compaction #62472).
Why it matters: Persisted contextTokens drives usage display, compaction thresholds, and related safeguards; a stale or collided value can make the runtime overly conservative or inconsistent with the active provider.
What changed: Use resolveContextTokensForModel({ cfg, provider, model, allowAsyncLoad: false }) in the auto-reply agent and follow-up runners, in persistInlineDirectives return value, and in resolveMemoryFlushContextWindowTokens (threading cfg + provider from the follow-up run). Follow-up runs now resolve providerUsed from agent meta when present, matching the main reply path.
What did NOT change: CLI updateSessionStoreAfterAgentRun already used provider-qualified resolution; gateway session row building and status logic were left as-is. The global context cache structure is unchanged; resolution goes through the existing resolveContextTokensForModel contract.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Closes [Bug]: Context token / context window can be persisted incorrectly when multiple providers share the same model id, causing wrong /status values and potentially premature compaction #62472
Related #
This PR fixes a bug or regression

Root Cause (if applicable)

Root cause: Mixed strategies: resolveContextTokensForModel correctly scans cfg.models.providers per provider, but several hot paths still used bare-model cache lookups or cached tokens keyed only by model id, so duplicate ids across providers could leak the wrong window into persisted session metadata.
Missing detection / guardrail: Unit coverage did not assert provider-qualified resolution for resolveMemoryFlushContextWindowTokens when two providers list the same model id with different limits.
Contributing context (if known): N/A

Regression Test Plan (if applicable)

Coverage level that should have caught this:
- Unit test
- Seam / integration test
- End-to-end test
- Existing coverage already sufficient
Target test or file: src/auto-reply/reply/reply-state.test.ts
Scenario the test should lock in: Two providers in config share one model id with different contextWindow values; resolveMemoryFlushContextWindowTokens returns the limit for the requested provider.
Why this is the smallest reliable guardrail: Exercises the same resolveContextTokensForModel path used by reply usage persistence and memory sizing without standing up a full channel run.
Existing test that already covers this (if any): Partial — resolveContextTokensForModel tests in src/agents/context.lookup.test.ts; this adds coverage at the memory-flush entry point.
If no new test is added, why not: N/A — test added.

User-visible / Behavior Changes

Session contextTokens and related usage / flush thresholds should match the active provider’s configured window when multiple providers reuse the same model id.

Diagram (if applicable)

N/A

Security Impact (required)

New permissions/capabilities? No
New network endpoints or trust boundaries? No

Testing

pnpm exec oxlint --type-aware on touched src/auto-reply/reply/*.ts files
pnpm test src/auto-reply/reply/reply-state.test.ts -t resolveMemoryFlushContextWindowTokens

greptile-apps · 2026-04-07T12:39:12Z

Greptile Summary

Fixes provider-qualified context window resolution across the memory-flush and preflight-compaction gates, inline-directive persist, and followup runners. Previously these paths used bare model-id lookups that could collide when multiple providers share a model id, writing the wrong contextTokens into session metadata and skewing compaction thresholds.

Confidence Score: 5/5

Safe to merge; the fix correctly threads provider context through all affected paths and adds targeted regression coverage.

All remaining findings are P2. The priority inversion for agentCfgContextTokens in the flush gate is a minor inconsistency that only affects rare deployments with explicit per-agent contextTokens overrides; the primary bug fix is correct and well-tested.

src/auto-reply/reply/memory-flush.ts — minor agentCfgContextTokens priority ordering

Prompt To Fix All With AI

This is a comment left during a code review.
Path: src/auto-reply/reply/memory-flush.ts
Line: 13-22

Comment:
**`agentCfgContextTokens` priority inconsistency with main run path**

In `agent-runner.ts` and `followup-runner.ts`, `agentCfgContextTokens` takes the highest priority (`agentCfgContextTokens ?? resolveContextTokensForModel(...)`), and in `directive-handling.persist.ts` it is passed as `contextTokensOverride` so it wins there too. Here the order is reversed — `resolveContextTokensForModel` wins, and `agentCfgContextTokens` only fires when no model is found in config. If an agent has an explicit `contextTokens` override smaller than the provider's configured window, the flush/compaction gate will use the larger model value and may not trigger when it should. Threading it as `contextTokensOverride` would restore consistency:

```suggestion
  return (
    resolveContextTokensForModel({
      cfg: params.cfg,
      provider: params.provider,
      model: params.modelId,
      contextTokensOverride: params.agentCfgContextTokens,
      allowAsyncLoad: false,
    }) ??
    DEFAULT_CONTEXT_TOKENS
  );
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "fix(sessions): provider-qualified contex..." | Re-trigger Greptile}

neeravmakwana · 2026-04-07T12:40:17Z

Review follow-up (commit `c248c9d`)

Greptile (inline on memory-flush.ts) — addressed
resolveMemoryFlushContextWindowTokens now passes agentCfgContextTokens into resolveContextTokensForModel as contextTokensOverride, matching agent-runner / followup-runner (leading agentCfgContextTokens ?? …) and directive-handling.persist (override inside resolveContextTokensForModel). The memory-flush / preflight-compaction gate therefore uses the same precedence: explicit per-agent contextTokens wins over provider/model config when both are set.

Regression test
Added prefers agent contextTokens override over the provider configured window in reply-state.test.ts (512k provider window vs 100k agent override → 100k).

Aisle Security (PR comment) — partial

Agent cap vs model window: The change above aligns flush gating with the agent cap; no separate “model wins over agent” path remains in this helper.
Non-finite / unbounded contextTokens: Sanitizing or clamping overrides and config-sourced limits inside resolveContextTokensForModel would be a broader contract change affecting many callers; it was not introduced by this PR. If we want a global clamp, that should be a dedicated follow-up with a shared constant and call-site audit.

All automated review items above are either implemented (Greptile + Aisle cap ordering) or explicitly scoped out (finite/clamp in core resolver) to avoid an unrelated refactor in this bugfix.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8b5bba51c6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@neeravmakwana

…w#62493) (thanks @neeravmakwana)

obviyus

Reviewed latest changes; landing now.

obviyus · 2026-04-09T11:55:57Z

Landed on main.

Landed source commit: 6e84928
Merge commit: 2645ed1

Thanks @neeravmakwana.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6e84928f1a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-09T12:02:39Z

-        sessionEntry?.contextTokens ??
-        DEFAULT_CONTEXT_TOKENS;
+        resolveContextTokensForModel({
+          cfg: queued.run.config,


Resolve follow-up context window from runtime config

This lookup uses queued.run.config even though the follow-up run itself is executed and persisted with runtimeConfig right below. If the runtime snapshot differs (for example after a hot config reload), contextTokensUsed can be derived from stale provider/model limits and then written into session usage, which skews /status reporting and memory-flush/compaction thresholds for that turn. Use the same runtimeConfig object here so context sizing matches the config actually used for the run.

Useful? React with 👍 / 👎.

…avmakwana

…avmakwana) * fix(sessions): provider-qualified context limits (#62472) * fix(sessions): honor agent context cap in memory-flush gate * refactor(sessions): unify context token resolution * fix: keep followup snapshot freshness on the active provider (#62493) (thanks @neeravmakwana) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>

@neeravmakwana

@neeravmakwana) * fix(sessions): provider-qualified context limits (openclaw#62472) * fix(sessions): honor agent context cap in memory-flush gate * refactor(sessions): unify context token resolution * fix: keep followup snapshot freshness on the active provider (openclaw#62493) (thanks @neeravmakwana) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>

@neeravmakwana

@neeravmakwana) * fix(sessions): provider-qualified context limits (openclaw#62472) * fix(sessions): honor agent context cap in memory-flush gate * refactor(sessions): unify context token resolution * fix: keep followup snapshot freshness on the active provider (openclaw#62493) (thanks @neeravmakwana) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>

@neeravmakwana

@neeravmakwana) * fix(sessions): provider-qualified context limits (openclaw#62472) * fix(sessions): honor agent context cap in memory-flush gate * refactor(sessions): unify context token resolution * fix: keep followup snapshot freshness on the active provider (openclaw#62493) (thanks @neeravmakwana) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>

@neeravmakwana

@neeravmakwana) * fix(sessions): provider-qualified context limits (openclaw#62472) * fix(sessions): honor agent context cap in memory-flush gate * refactor(sessions): unify context token resolution * fix: keep followup snapshot freshness on the active provider (openclaw#62493) (thanks @neeravmakwana) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>

@neeravmakwana

@neeravmakwana) * fix(sessions): provider-qualified context limits (openclaw#62472) * fix(sessions): honor agent context cap in memory-flush gate * refactor(sessions): unify context token resolution * fix: keep followup snapshot freshness on the active provider (openclaw#62493) (thanks @neeravmakwana) --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>

neeravmakwana mentioned this pull request Apr 7, 2026

[Bug]: Context token / context window can be persisted incorrectly when multiple providers share the same model id, causing wrong /status values and potentially premature compaction #62472

Closed

openclaw-barnacle Bot added the size: S label Apr 7, 2026

greptile-apps Bot reviewed Apr 7, 2026

View reviewed changes

Comment thread src/auto-reply/reply/memory-flush.ts

obviyus force-pushed the fix/context-tokens-provider-62472 branch from c248c9d to 8b5bba5 Compare April 9, 2026 11:16

chatgpt-codex-connector Bot reviewed Apr 9, 2026

View reviewed changes

Comment thread src/auto-reply/reply/followup-runner.ts

obviyus self-assigned this Apr 9, 2026

obviyus force-pushed the fix/context-tokens-provider-62472 branch from 8b5bba5 to 6b8859c Compare April 9, 2026 11:50

neeravmakwana and others added 4 commits April 9, 2026 17:23

fix(sessions): provider-qualified context limits (openclaw#62472)

bfb0798

fix(sessions): honor agent context cap in memory-flush gate

65b9626

refactor(sessions): unify context token resolution

d4c5f05

fix: keep followup snapshot freshness on the active provider (opencla…

6e84928

…w#62493) (thanks @neeravmakwana)

obviyus force-pushed the fix/context-tokens-provider-62472 branch from fcc533d to 6e84928 Compare April 9, 2026 11:54

obviyus approved these changes Apr 9, 2026

View reviewed changes

obviyus merged commit 2645ed1 into openclaw:main Apr 9, 2026
23 of 26 checks passed

This was referenced Apr 9, 2026

Bug: contextTokens/status can use wrong context window when same model id exists across providers #49355

Closed

Status summary should resolve context windows by provider-qualified model key #36003

Closed

chatgpt-codex-connector Bot reviewed Apr 9, 2026

View reviewed changes

github-actions Bot mentioned this pull request Apr 9, 2026

📡 Upstream Digest — 2026-04-09 13:04 UTC curtismercier/openclaw-mods#520

Open

clawsweeper Bot mentioned this pull request May 4, 2026

feat(config): add llm shorthand config key for custom model providers #51603

Closed

3 tasks

clawsweeper Bot mentioned this pull request May 12, 2026

fix(reply): gate preflight compaction fast-path on token threshold (#63892) #64384

Open

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(sessions): provider-qualified context limits (#62472)#62493

fix(sessions): provider-qualified context limits (#62472)#62493
obviyus merged 4 commits into
openclaw:mainfrom
neeravmakwana:fix/context-tokens-provider-62472

neeravmakwana commented Apr 7, 2026

Uh oh!

greptile-apps Bot commented Apr 7, 2026

Uh oh!

Uh oh!

neeravmakwana commented Apr 7, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

obviyus left a comment

Uh oh!

Uh oh!

obviyus commented Apr 9, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Apr 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

neeravmakwana commented Apr 7, 2026

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Root Cause (if applicable)

Regression Test Plan (if applicable)

User-visible / Behavior Changes

Diagram (if applicable)

Security Impact (required)

Testing

Uh oh!

greptile-apps Bot commented Apr 7, 2026

Greptile Summary

Confidence Score: 5/5

Uh oh!

Uh oh!

neeravmakwana commented Apr 7, 2026

Review follow-up (commit c248c9d)

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

obviyus left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

obviyus commented Apr 9, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Review follow-up (commit `c248c9d`)