fix(gateway): a steer/interrupt-triggered queued terminal turn gets its own ledger id

[nftpoetrist]

What does this PR do?

Closes the one branch the queued-lane ledger saga (#106312/#106316, merged earlier today) didn't reach: a queued follow-up triggered by a leftover /steer or a bare control interrupt_message (no originating queued event) still collides on the same delivery-ledger obligation id as an earlier reply in the chain.

Problem

gateway/run_turn.py::_run_agent_queued_followup derives next_inbound_id — the id that keys the follow-up's own delivery-ledger obligation — only inside the if pending_event is not None: branch. When the follow-up instead comes from result["pending_steer"] or a bare interrupt_message (pending_event is None), next_inbound_id stays at its None default.

That None flows into queued_terminal_inbound_id, which the caller in gateway/run.py reads back with a truthiness check:

_terminal_inbound = agent_result.get("queued_terminal_inbound_id")
if _terminal_inbound:
    event.ledger_message_id = str(_terminal_inbound)

None is indistinguishable from "absent" here, so event.ledger_message_id is never overridden and the outer final send falls back to the id of the event that opened the chain. If this steer/interrupt-triggered terminal reply happens to carry the same text as an earlier reply in the same chain, both compute the identical obligation id — the earlier (possibly flood-refused) row is silently overwritten and marked delivered, and is never redelivered. This is exactly the bug the saga just fixed, reached from the one branch it didn't cover.

Fix

When pending_event is None, mint a synthetic id (uuid4, prefixed steer:) distinct from any real transport id. There's no natural identity to derive one from for a synthesized continuation — this mirrors the existing keyless-turn-owner fallback already used elsewhere in the same file (_run_agent_prepare_turn's owner id, which falls back to uuid.uuid4() when event.message_id is absent).

Testing

• Added test_a_steer_triggered_terminal_turn_gets_its_own_inbound_id, mirroring the saga's own test_a_deeper_chain_keeps_the_innermost_inbound_id pattern: asserts the inbound id is truthy, that it feeds into queued_terminal_inbound_id, and that two separate occurrences get different ids (not a deterministic/reusable value).
• Mutation-verified: reverting gateway/run_turn.py drops exactly the new test (9 pass, 1 fail — assert None, reproducing the bug directly); reapplying restores green (10 passed).
• uv run --frozen python -m pytest tests/gateway/test_queued_final_ledger.py tests/gateway/test_queued_native_image_session_key.py tests/gateway/test_multiplex_busy_input_mode.py -q → 32 passed (checked sibling queued-lane suites for interaction regressions — none).
• ruff check clean on both changed files.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Checklist

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)

[Enough1122]

AI code review — automated review for reference; please use your judgment.

Summary

In gateway/run_turn.py::_run_agent_queued_followup, a steer/interrupt-triggered terminal turn with no originating queued event now mints a unique steer:<uuid> inbound id instead of leaving next_inbound_id=None (which silently skipped the ledger override and fell back to the chain-opening event, risking same-text obligation-id collision — the branch the #106312/#106316 saga missed). Test asserts truthy, distinct-across-occurrences ids.

Findings

No blocking issues.

• Root-cause chain is right: None failed the truthiness check at the call site, so the override was skipped silently. Minting a fresh uuid per occurrence guarantees distinct ledger identity; steer: prefix keeps it greppable and disjoint from real transport ids.
• Import placement: import uuid is function-local in the else branch — stdlib import is cheap/cached; fine, though module top-level would be more conventional. Cosmetic.
• Test covers the exact regression (second steer drain gets a different id). Good.

Verdict

Looks good. No concerns.