GeezerVison

The native engine now translates the existing session-level noContextFiles limit into an InstructionOptions.off flag for synchronous and asynchronous tool execution. The nearby-instruction reader returns no instruction text or claims when that flag is set, while the read tool still returns the requested file. The same options are passed to orientation generation, although its existing instruction-file discovery does not consume off; mention expansion also remains unchanged.

Changes

Sequence diagram

sequenceDiagram
    participant Engine as Native engine
    participant Limits as Session run limits
    participant Read as Read tool
    participant Instructions as Nearby instruction reader
    Engine->>Limits: Get noContextFiles
    Limits-->>Engine: Session policy
    Engine->>Read: Execute with instructions.off
    Read->>Instructions: readInstructions(file, context)
    alt off is true
        Instructions-->>Read: Empty text and loaded claims
    else off is absent or false
        Instructions-->>Read: Existing nearby instruction behavior
    end
    Read-->>Engine: Requested file output

• Files reviewed: 5 · 24 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: 251,097 tokens in (211,712 from the provider's cache) · 1,985 out
• Readers: 4 files went to readers instead of the review pass whole: 4 code files by gpt-6-luna (the review pass gets their summaries and the hunks of their risk spots; 1 call, up to 3 at once, 8,786 tokens counted). Their summaries are hints, not findings. 1 code file no reader could read went to the review pass whole (the reader gave no usable answer for this file).
• Stacked on #133: only this pull request's own commits since #133's head 785277d.

Performance (perf-attrib, base 785277d against head 9d72b91)

perf-attrib: baja: tui.after.heap.protected -1

base 785277dbae21b3f364b365ed6e78e02fd6c08280 (785277db) vs cand 9d72b91b8cdaa40cb392ff84d087d7d56a0cde16 (9d72b91b), median of 2 faux-engine session(s) per side; flagged = beyond the noise measured on 18 base-vs-base runs (x1.5).

JS heap bytes by constructor that rose:

• tui.rest.ctor.object:ModuleRecord.bytes: 5502928 -> 5503025 (+97)
• tui.rest.ctor.native:Blob.bytes: 1739300 -> 1739389 (+89)

Modules loaded at start: 230 -> 230, 4631407 -> 4631504 bytes of chunks; chunk sizes only in cand: [250693, 181666], only in base: [250635, 181627] (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): 54.7 / 57.1 / peak 73.1 MB, 2056 B per session.