fix(gateway): preserve queued follow-up media delivery

[vKongv]

What does this PR do?

When a second inbound message queues behind an in-progress gateway turn, the queued resend path previously replayed only the text body. MEDIA-backed attachments from send_asset could be dropped silently. This routes queued first-response resends through a text-plus-media delivery helper instead of a plain adapter text send.

Related Issue

Fixes #

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• gateway/run.py — route queued first-response resends through _deliver_queued_first_response; preserve MEDIA attachments the queued path can replay natively
• tests/gateway/test_tts_media_routing.py — regression test for queued follow-up media delivery

How to Test

1. uv run pytest -o addopts= tests/gateway/test_tts_media_routing.py -k queued -q
2. Trigger a gateway turn that returns a response with MEDIA: attachments, then queue a second inbound before delivery completes — confirm images are delivered on the queued resend path

Checklist

Code

• 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)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

For New Skills

N/A — delete this section otherwise.

Screenshots / Logs

N/A

[alt-glitch]

Likely duplicate of #18546 — same bug: gateway queue drain loop skips extract_media() for queued follow-up responses, dropping MEDIA file delivery. #19011 was already flagged as a duplicate of #18546 for the same issue.

[teknium1]

Thanks for narrowing this to the queued first-response resend path. The defect is present on current main: gateway/run.py:19724-19728 directly sends first_response, while native attachment routing lives in _deliver_media_from_response() at gateway/run.py:13097-13216.

Problems

• The added broad cleanup at gateway/run.py:2188 strips every remaining MEDIA: token after extract_media(). That conflicts with the base contract: protected code/inline-code spans and unsupported or unvalidated tags are deliberately preserved (gateway/platforms/base.py:3623-3675, 1460-1481).
• The new tests call the helper directly, rather than taking the queued resend branch at gateway/run.py:19718-19730; they do not verify the actual integration point.

Suggested changes

• Drop the broad regex and retain the base extractor's cleaned text semantics.
• Add a queued-follow-up integration test, including attachment routing and a protected MEDIA: example.

Automated hermes-sweeper review.

[vKongv]

Thanks for the heads-up on #18546 / #19011 — same gap on the queued first-response resend path.

This PR keeps that path, but routes delivery through the existing text + media helper so attachments still get the normal allowlist / type-aware routing instead of a one-off extract-and-send loop. Also dropped the broad leftover MEDIA: strip after extract_media() (it was wiping protected code/inline examples), and added a regression that hits the real queued resend branch in _run_agent.

Rebased on current main. Happy to close/consolidate if maintainers want a single landing PR — otherwise this should be ready for another look.

[alt-glitch]

This was generated by AI during triage.

Current-head correction: #25119 predates the other open queued-media PRs. It shares the queued first-response delivery outcome but uses a distinct helper shape, so this is a competing implementation rather than a PR-to-issue duplicate. Related: #18546, #46223, and #51805.

[SemonCat]

Opened a small incremental follow-up directly on top of this branch: pebble-tech#56.

It covers the one remaining gap in the queued-follow-up flush: the "already streamed" branch (where streaming already delivered the first response's text) never ran the MEDIA: upload pass, so an attachment carried by that same response was silently dropped — streaming only ever delivers text. The fix reuses _deliver_media_from_response (the same helper _deliver_queued_first_response already calls for its media half) via a synthetic event; no new helper, no change to the direct-send branch this PR already fixes.

Also closing #77732 — a duplicate that a triage pass flagged as part of this same family — in favor of consolidating that one missing piece here rather than competing with your work.

[teknium1]

Closing as superseded — with credit: #82162 (salvage of #71031) landed on main and fixes queued follow-up MEDIA delivery for both branches. Your #25119 commits were consolidated into that PR branch with attribution preserved before the salvage, so your work is part of what shipped. Thanks!