pr_queue_snapshot: codex bot's PR-level review comment bucketed as 'new human feedback', not bot feedback

[kyle-sexton]

Observed

Cycle 14 of the babysit-prs loop dispatched a worker to #681 partly on the strength of the snapshot line:

```
new human feedback: chatgpt-codex-connector ### 💡 Codex Review Here are some automated review suggestions...
```

The dispatched worker fetched all three comment surfaces (issue comments, review comments, GraphQL reviewThreads) and found no human comment or human thread on the PR at all — only bot authors. Same pattern showed up on #665 and #676 in the same snapshot (new human feedback: chatgpt-codex-connector ...Codex Review, and separately new human feedback: kyle-sexton COMMENTED — an empty-bodied review event, not prose feedback).

Root cause (suspected)

chatgpt-codex-connector's summary comment appears to land as a top-level PR issue comment (or a COMMENTED-state PR review body) rather than an inline reviewThreads entry. The snapshot's bot/human classifier for the "new feedback" delta appears to key off comment placement (issue-comment vs. review-thread) rather than author type (__typename == "Bot" / [bot] suffix) for this feedback category — so a clearly bot-authored top-level comment gets bucketed as "new human feedback" instead of "new blocking feedback."

This is the same class of author-vs-placement classification gap as #534 (closed) but appears to be a surviving instance in the "new feedback" delta lines specifically, separate from unresolvedThreads/blockers which classify correctly (those show chatgpt-codex-connector under blocking feedback, not blocking human feedback, for the same PR).

Impact

Low-moderate: causes worker dispatch briefs to claim "new human feedback exists" when it doesn't, sending a worker on an unnecessary verification detour (harmless, but wastes a tool round and can read as untriaged human input needing a reply when none exists).

Suggested fix

Classify "new ... feedback" delta lines by the same author-type check already used for blockers/unresolvedThreads (structural __typename == "Bot" or [bot] login suffix), not by comment placement (issue-comment vs. reviewThread).

Filed by the babysit-prs autonomous loop (self-observation discipline), cycle 14, 2026-07-20.

Work-class: C3 (bug-fix-shaped) — attended triage 2026-07-23, operator-ratified. 🤖

[kyle-sexton]

This was generated by AI on operator direction.

Superseded — this item was already ratified and should not have been queued.

The earlier first-drain ratification request on this issue was raised in error. The admission gate dispatches a C3 bug-fix-shaped item when it carries either its own recorded ratification or a completed first drain — and this issue already carried recorded ratification in its body from attended triage on 2026-07-23. The drain cycle applied the blanket first-drain rule without first reading bodies for an existing ratification, so it queued an item that was already admitted.

Restored to the autonomous frontier: agent-ready re-applied, needs-human removed. No human action is required, and this issue no longer appears in the attended queue's ratify view.

[kyle-sexton]

This was generated by AI during the work-loop admission gate.

Classification

• Work class: C3 scoped, bug-fix-shaped — pr_queue_snapshot mis-buckets the codex bot's
  PR-level review comment as new human feedback instead of bot feedback; a bounded fix to
  existing classification logic, not a new capability.
• Type: Bug (tracker's existing type).

Intended dispatch

Queued for the first-drain C3 ratification gate — first_drain_complete is still false, so
this bug-fix-shaped C3 candidate needs its own recorded ratification before dispatch. Ratify by
replying on this comment via /work-items:attend-queue, or it clears automatically once
first_drain_complete flips true.

[kyle-sexton]

This was generated by AI during the work-loop admission gate, on the operator's behalf.

Ratification

Operator ratified this item's C3 bug-fix-shaped classification and dispatch, direct in the
/work-items:work-loop session (standing autonomous-lane authorization, explicit "take on up to
5 items" instruction). Flipping to autonomous-eligible; re-entering the worker loop's frontier for
dispatch this cycle.