Move run completion coordination out of the session host

[ThomasK33]

What to build

Move run-completion coordination out of src/host/hostMain.ts into a dedicated module. The new module should coordinate marker/sentinel creation, postamble echo registration, waiter resolution, timeout behavior, exit resolution, and run_complete event appends while preserving the existing sentinel scanner module.

Acceptance criteria

• hostMain delegates run-completion coordination to a dedicated module while still owning PTY and RPC wiring.
• Timeout behavior keeps sentinel/postamble registrations active so eventual internal bytes are still hidden from artifacts.
• Session exit resolves pending run-completion waiters the same way it does today.
• Existing run-completion sentinel unit tests and run command integration tests still pass, especially tests that assert sentinel text does not leak into artifacts.

Blocked by

None - can start immediately.

[ThomasK33]

This was generated by AI during triage.

Agent Brief

Category: enhancement
Summary: Move run-completion coordination out of the session host while preserving existing run behavior and artifact cleanliness.

Current behavior:
The session host currently owns both PTY/RPC orchestration and run-completion coordination. That coordination includes marker creation, short sentinel generation, postamble echo registration, pending waiter tracking, timeout behavior, exit resolution, and appending structured run_complete events. The sentinel scanner and postamble echo sanitizer already exist as dedicated run-completion helpers.

Desired behavior:
Run-completion coordination should be owned by a dedicated host-side module. The session host should still own PTY lifecycle, RPC wiring, command dispatch, and event-log ingestion order, but should delegate run-completion-specific state transitions and bookkeeping to the new module.

The behavior must remain unchanged for callers: waited runs still return the same accepted/completed/timed-out result shape, timeouts still leave internal sentinel/postamble registrations active so eventual internal bytes stay out of persisted artifacts, and session exit still resolves pending run-completion waits the same way it does today.

Key interfaces:

• RunResult for waited and no-wait runs — the public result shape must not change.
• input_run and run_complete event records — event payloads and sequence semantics must remain compatible with existing replay and export code.
• RunCompletionSentinelScanner and RunCompletionPostambleEchoSanitizer — keep these as the existing low-level sentinel/echo filtering helpers rather than replacing their scanning behavior.
• Event-log append behavior for output and run_complete records — preserve serialized PTY ingestion so artifacts remain reproducible from the event log.
• Session exit handling — pending run-completion waiters must resolve as exited without turning ordinary PTY exit into a run timeout.

Acceptance criteria:

• The session host delegates run-completion coordination to a dedicated module while retaining PTY lifecycle, RPC wiring, and command dispatch ownership.
• Waited run calls preserve the existing JSON result fields and semantics for completed, timed-out, and exited runs.
• Timeout behavior keeps sentinel and postamble echo registrations active after the caller receives a timeout, so later internal completion bytes are still hidden from event-log output, snapshots, casts, and other artifacts.
• Session exit resolves pending run-completion waiters the same way it does before the refactor.
• Structured run_complete events are still appended exactly when an active run-completion sentinel is observed, with the correct marker and input-run sequence relationship.
• Existing run-completion sentinel unit tests and run command integration tests pass, including coverage that asserts sentinel/postamble text does not leak into artifacts.
• Typecheck and lint pass for the extracted module and updated host wiring.

Out of scope:

• Changing the public CLI JSON contract for run.
• Changing protocol schemas, event-log record shapes, or replay/export semantics except where tests require preserving existing behavior.
• Rewriting the existing sentinel scanner or postamble echo sanitizer algorithms.
• Adding new run modes, shell-specific behavior, or renderer behavior.

Triage validation:
The focused validation run passed before marking this ready: sentinel unit tests and run command integration tests both passed.