GeezerVison

The bridge now checks for an active mirrored turn both before and after ensuring the chat's session, returning a wait-or-stop notice instead of submitting another prompt. Subscription setup reads the event-log boundary before checking running-turn history, then chooses the earlier replay cursor. The patch also adds focused regression tests and updates release notes, channel documentation, and the generated card revision; daemon submission, history, and replay contracts remain unchanged.

Changes

Sequence diagram

sequenceDiagram
    participant Chat as Telegram chat
    participant Bridge as Channel bridge
    participant Client as Daemon client
    participant Daemon
    Bridge->>Client: loadWindow(sessionId, 1)
    Client->>Daemon: session.events
    Daemon-->>Client: Event-log boundary
    Bridge->>Client: session.history(limit: 1)
    Client->>Daemon: Read latest turn
    Daemon-->>Client: Turn history
    Bridge->>Client: subscribe(min(boundary, running start - 1))
    Client->>Daemon: client.subscribe
    Daemon-->>Bridge: Replay followed by live events
    Chat->>Bridge: Prompt during mirrored turn
    Bridge-->>Chat: Wait for the answer or /stop

• Files reviewed: 6 · 50 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: 291,560 tokens in (253,312 from the provider's cache) · 2,349 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).
• Stacked on #146: only this pull request's own commits since #146's head b3a2066.

Performance (perf-attrib, base 1aed947 against head 8346c30)

perf-attrib: igual: nothing above the measured noise

base 1aed947efa71e3694ca6a5175cd0cb796b421fc6 (1aed947e) vs cand 8346c308c7e1e22a7da32d1986b40b86aaf71e16 (8346c308), 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.Function: 44528 -> 44529 (+1)
• tui.rest.heap.type.Function: 44524 -> 44525 (+1)

JS heap bytes by constructor that rose:

• tui.rest.ctor.native:Blob.bytes: 1743266 -> 1743695 (+429)
• tui.rest.ctor.object:ModuleRecord.bytes: 5504809 -> 5504845 (+36)

Modules loaded at start: 230 -> 230, 4633288 -> 4633324 bytes of chunks; chunk sizes only in cand: [1009178], only in base: [1009142] (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 / 56.7 / peak 73.1 MB, 2038 B per session.