GeezerVison

The existing unreleased notes become the notes for v0.0.14-beta.1, with a new summary highlighting the release’s user-visible improvements. A separate v0.0.14 entry repeats that summary and states that it contains the same code as the beta, following the preceding release’s format. Existing bullets are retained, with several reordered, and Unreleased remains available for future changes.

Changes

• Files reviewed: 1 · 18 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: 13,970 tokens in (5,888 from the provider's cache) · 495 out
• Readers: 1 code file 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).
• Stacked on #140: only this pull request's own commits since #140's head d608ea3.

This increment is a Computer Use lifecycle fix rather than the changelog-only release described in the pull request. Previously, a sweep could release an idle session while another operation still needed native resources. Sweeping now waits for global quiescence, and session starts are counted through a finally block so both successful and failed starts trigger another sweep. Session authorization, per-call expiry checks, and the native release implementation remain unchanged.

Changes

Sequence diagram

sequenceDiagram
    participant Caller
    participant Sessions as Computer sessions
    participant Native as Accessibility backend
    Caller->>Sessions: Start session
    Sessions->>Sessions: Increment pending starts
    Sessions->>Native: Resolve or identify target
    Sessions->>Sessions: Register session if successful
    Sessions->>Sessions: Finally decrement pending starts and sweep
    alt Sessions, operations, or captures still hold resources
        Sessions->>Sessions: Schedule next sweep
    else All resource holders are quiet
        Sessions->>Native: Release expired sessions' resources
    end

• Files reviewed: 2 · 32 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: 172,149 tokens in (140,160 from the provider's cache) · 1,605 out
• Readers: 2 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 d608ea3 against head dee18c8)

perf-attrib: baja: tui.after.heap.protected -1; tui.rest.fp_mb -5.2MB; total.rest.fp_mb -5.2MB

base d608ea3b230fa30b4d4bdc528a4bcb786ef25629 (d608ea3b) vs cand dee18c87e34bc15d8ae6b4a1dc68b9d522f2092e (dee18c87), median of 2 faux-engine session(s) per side; flagged = beyond the noise measured on 18 base-vs-base runs (x1.5).

JS object types that rose (count):

• tui.after.heap.type.Object: 38318 -> 38320 (+2)
• tui.after.heap.type.StructureChain: 91 -> 93 (+2)
• tui.rest.heap.type.Object: 38326 -> 38328 (+2)
• tui.rest.heap.type.StructureChain: 90 -> 92 (+2)

JS heap bytes by constructor that rose:

• tui.rest.ctor.native:Blob.bytes: 1741200 -> 1741846 (+646)
• tui.rest.ctor.object:Object.bytes: 1412688 -> 1412768 (+80)
• tui.rest.ctor.array:Array_?0?.bytes: 35938 -> 35954 (+16)

Modules loaded at start: 230 -> 230, 4631481 -> 4631481 bytes of chunks (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): 54.8 / 57.2 / peak 73.2 MB, 2056 B per session.