GeezerVison

The patch addresses two Telegram remote-control races on top of the parent pull request. Phone submissions explicitly select queued delivery so a concurrent computer turn cannot consume them as steering input. Additional chats replay the running turn through paginated event reads instead of resetting the shared subscription cursor; the existing bridge buffers and deduplicates that replay. Protocol schemas, kernel queue behavior, and normal client subscription handling remain unchanged.

Changes

Sequence diagram

sequenceDiagram
    participant Chat as Joining Telegram chat
    participant Bridge as Channel bridge
    participant Host as Channels host
    participant Client as Shared daemon client
    participant Daemon
    Chat->>Bridge: Pair or select session
    Bridge->>Host: watch(sessionId)
    Host->>Client: loadWindow and session.history
    Client->>Daemon: Read end cursor and running turn
    alt First chat following session
        Host->>Client: subscribe(from)
        Client->>Daemon: Replay and subscribe
        Daemon-->>Bridge: Session events via client
    else Session already followed
        loop Up to 50 pages
            Host->>Client: loadAfter(after, 200)
            Client->>Daemon: session.events
            Daemon-->>Host: Event page via client
            Host-->>Bridge: onEvent for replayed events
        end
    end
    Host-->>Bridge: Watch result
    Bridge->>Bridge: Sort and deduplicate buffered events
    Bridge-->>Chat: Prompt, progress and approvals

• Files reviewed: 5 · 56 changed lines · estimated review effort 2/5, about 15 minutes
• Profile: normal · at most 8 inline comments
• Verification: every finding was checked again by a separate pass (1 candidate judged)
• Cost: 417,797 tokens in (309,888 from the provider's cache) · 3,438 out
• Readers: 5 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 a71641f.

Performance (perf-attrib, base 8346c30 against head 22d35bc)

perf-attrib: igual: nothing above the measured noise

base 8346c308c7e1e22a7da32d1986b40b86aaf71e16 (8346c308) vs cand 22d35bcd6d29d479aea924a923d0fe01b4183328 (22d35bcd), 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.native:Blob.bytes: 1743695 -> 1744090 (+395)
• tui.rest.ctor.object:ModuleRecord.bytes: 5504845 -> 5505128 (+283)
• tui.rest.ctor.code:JSModuleEnvironment.bytes: 104208 -> 104224 (+16)

Modules loaded at start: 230 -> 230, 4633324 -> 4633607 bytes of chunks; chunk sizes only in cand: [1009461], only in base: [1009178] (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): 54.5 / 57.1 / peak 73.4 MB, 2030 B per session.