Repro

Discovered via the GitHub roundtrip integration harness (TestRoundtrip_ReplyToRemoteComment).

1. A remote (GitHub PR) review comment exists on a line.
2. Run crit pull — the comment is imported into the local review file with a fresh local string ID and github_id set.
3. Run crit comment --reply-to <local-id> "ack" — the reply is appended locally under the imported comment's replies.
4. Run crit push.

Expected: a new GitHub PR review comment is posted with in_reply_to: <parent github_id>, so it appears as a threaded reply.

Actual: crit push posts nothing. Remote comment count is unchanged.

=== RUN   TestRoundtrip_ReplyToRemoteComment
    roundtrip_integration_test.go:189: want 1 new remote item, got 0:
          id=3180419068 parent=0 path=sample.go line=19 body=please address
--- FAIL: TestRoundtrip_ReplyToRemoteComment (11.28s)

Likely cause

crit push probably treats the imported root as already-synced (it has a github_id) and only walks new top-level comments, skipping fresh local replies whose parent has a github_id. Reply-push logic likely only fires for replies whose parent was authored locally.

Sandbox

crit-md/crit-roundtrip-sandbox. Reproduces deterministically against PR #1 with the harness in crit.pr-roundtrip-harness branch (test file: roundtrip_integration_test.go::TestRoundtrip_ReplyToRemoteComment).

Scope

Test-only investigation — production fix tracked here, not in the harness branch.