GeezerVisonThe World Wide Web, restored to 800 pixels of glory. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Home|About|Gaming cabinet|Archives|How it works|Starter sites | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
GeezerVison is showing github.com — (1/9) Claude Code compacts at 200K unless a per-model point raises it by Obed0101 · Pull Request #128 · MendCode/MendCode Conversation
|
|
MendCode: @Obed0101's MendCode worked on this pull request on one of its devices; released 2026-11-10 18:19 UTC. |
All reactions
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Obed0101
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
MendCode review
No findings · nothing in this change was confirmed as a problem
head 2c95e59b7f3a
Merge risk: 🟢 Low · The change adjusts one numeric fallback in an existing shared helper, with no schema, persistence, dependency, or call-structure changes. A regression would primarily affect when Claude Code compacts and the compaction point displayed by its existing reporting paths.
Change: Bug fix · Review effort: 1/5, about 5 minutes
The change caps Claude Code's globally configured compaction point at 200,000 tokens while preserving lower global values and higher per-model or environment overrides. No introduced defects were found after reviewing the implementation, callers, validation, and affected tests; risk is low because the change is narrowly scoped. Tests were not executed in this read-only review, and the supplied performance measurements report no regression above noise.
Readers: 6 code files no reader could read went to the review pass whole (the diff is small: one reader call costs more than the readers' share of this review's cap, so the review pass reads it).
Walkthrough
Claude Code previously inherited global compaction settings even when those settings targeted long-window models. The helper now limits the global fallback to 200,000 tokens without changing the precedence of environment and per-model overrides or the existing feature guards. Existing process configuration, context reporting, and model reporting continue to consume the same helper; native-engine compaction is unchanged.
Changes
| Layer / files | Summary |
|---|---|
Claude Code compaction policy packages/daemon/src/claude-models.ts |
Caps the global compaction fallback at 200,000 tokens while preserving explicit overrides and existing opt-out behavior. |
Regression tests packages/daemon/test/context-truth.test.ts, packages/daemon/test/models.test.ts |
Updates expectations for capped global settings and checks lower global values, higher per-model settings, and environment precedence. |
Documentation and generated metadata CHANGELOG.md, docs/claude-code.md, packages/daemon/cards/cards.generated.json |
Documents the new fallback policy and updates the generated cards revision without changing card contents. |
Pre-merge checks: ⚠️ 1 warning · ✅ 2 passed
| Check | Status | Explanation |
|---|---|---|
| Title | ✅ Passed | The title accurately identifies the 200K cap and the retained ability to raise it per model. |
| Description | The description explains the intended behavior and reports a clean head typecheck, but its full-suite result applies to the top of the stack rather than this isolated head. Its claim of no per-turn work is also imprecise: packages/engine-claude/src/engine.ts:715 invokes the environment callback when checking whether a connection can be reused. | |
| Out of scope | ✅ Passed | The implementation, tests, documentation, and generated revision update are consistent with the stated compaction-policy change. |
Review details
- Files reviewed: 6 · 17 changed lines · estimated review effort 1/5, about 5 minutes
- Profile: normal · at most 8 inline comments
- Verification: the review pass raised no candidate findings, so no separate pass was needed
- Cost: 181,848 tokens in (153,344 from the provider's cache) · 1,686 out
- Readers: 6 code files no reader could read went to the review pass whole (the diff is small: one reader call costs more than the readers' share of this review's cap, so the review pass reads it).
Performance (perf-attrib, base c0516da against head 2c95e59)
perf-attrib: igual: nothing above the measured noise
base c0516da2078ccf2cdcd2c6688e567b10b48ac29c (c0516da2) vs cand 2c95e59b7f3a05988396170baea350d411e8a2a4 (2c95e59b, cached build), median of 2 faux-engine session(s) per side; flagged = beyond the noise measured on 18 base-vs-base runs (x1.5).
| metric | base | cand | diff |
|---|---|---|---|
| tui.after.fp_mb | 57.4 | 57.1 | -0.3 |
| tui.after.peak_mb | 73.2 | 73.1 | -0.1 |
| tui.ready.fp_mb | 54.9 | 54.9 | 0 |
| tui.ready.peak_mb | 73.2 | 73.1 | -0.1 |
| tui.rest.fp_mb | 56.9 | 56.5 | -0.4 |
| tui.rest.peak_mb | 73.2 | 73.1 | -0.1 |
| tui.after.rss_mb | 136.5 | 135.8 | -0.7 |
| tui.ready.rss_mb | 114 | 114.1 | +0.1 |
| tui.rest.rss_mb | 136.5 | 135.8 | -0.7 |
| tui.after.heap.extra_b | 5742500 | 6778245 | +1035745 |
| tui.after.heap.objects | 238306 | 240813 | +2507 |
| tui.after.heap.protected | 19 | 19 | 0 |
| tui.after.heap.size_b | 23149012 | 24445397 | +1296385 |
| tui.ready.heap.extra_b | 2871766 | 2895000 | +23234 |
| tui.ready.heap.objects | 216397 | 216466 | +69 |
| tui.ready.heap.protected | 18 | 18 | 0 |
| tui.ready.heap.size_b | 19241750 | 19273455 | +31705 |
| tui.rest.heap.extra_b | 3774179 | 4668849 | +894670 |
| tui.rest.heap.objects | 235291 | 237655 | +2364 |
| tui.rest.heap.protected | 18 | 18 | 0 |
| tui.rest.heap.size_b | 20924979 | 22060450 | +1135471 |
| tui.rest.js_code_b | 4924995 | 5878816 | +953821 |
| tui.rest.js_live_b | 19202727 | 19237774 | +35047 |
| tui.cpu_ms.after | 5922 | 5928 | +6 |
| tui.cpu_ms.ready | 358 | 353 | -5 |
| tui.cpu_ms.rest | 534 | 523 | -11 |
| binary.bytes | 72290528 | 72290528 | 0 |
| disk.bytes | 4106 | 4113 | +7 |
| disk.bytes_per_session | 2053 | 2057 | +4 |
JS heap bytes by constructor that rose:
- tui.rest.ctor.native:Blob.bytes: 1738553 -> 1738720 (+167)
- tui.rest.ctor.closure:G.bytes: 2721 -> 2753 (+32)
- tui.rest.ctor.object:ModuleRecord.bytes: 5501477 -> 5501493 (+16)
Modules loaded at start: 230 -> 230, 4629956 -> 4629972 bytes of chunks; chunk sizes only in cand: [5932], only in base: [5916] (chunk names are hashes: the list says which changed in size, not by what name)
Reference (skill, 2026-11-03, real model): footprint 64 ready / 66 after / peak 80 MB, 2.6 KB per session. This run (faux engine, whole process tree): 54.9 / 57.1 / peak 73.1 MB, 2057 B per session.
Files (6)
CHANGELOG.mddocs/claude-code.mdpackages/daemon/cards/cards.generated.jsonpackages/daemon/src/claude-models.tspackages/daemon/test/context-truth.test.tspackages/daemon/test/models.test.ts
Comment @mendcode review to review again · @mendcode pause / resume · reply to a finding with explain, fix or learn · settings in .mendcode/review.yaml
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
All reactions
Uh oh!
There was an error while loading. Please reload this page.
Reviewers
No reviewsUh oh!
There was an error while loading. Please reload this page.
Assignees
No one assignedLabels
Projects
Milestone
No milestoneDevelopment
Successfully merging this pull request may close these issues.
Uh oh!
There was an error while loading. Please reload this page.
1 participant
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.