[grug] Living stack status: in-progress note, commit-range disclosure, run-config section

[quadseven]

What

Make the Elder/Teller comment stack show LIVE review state, FLINT-style:

1. While a re-review runs, the stack comment shows a "Grug reading new markings - wait" note instead of stale results.
2. Every review run discloses its commit range explicitly: "Reviewing changes from and between and " (the Living Hunt delta machinery from [grug] Incremental per-push re-review: commit-range scoping, auto-pause, resume commands #557/feat(elder): Living Hunt foundation - delta review since last head (#557) #644 already computes this - surface it).
3. Disclose the run posture in a collapsed section: tier used (single-arm/deep), settle window, arms, model labels (codenames only).

Why

FLINT's dynamic status is a large part of why its reviews FEEL responsive and trustworthy: you always know whether results are current, what range they cover, and what configuration produced them. Grug computes all three facts already and shows none of them.

Acceptance criteria

• Pushing to an open PR flips the stack comment to an in-progress note. NOT DONE - moved to feat(grug): board shows a re-review is running, without the two ways that broke it #819. Built, reviewed, removed before shipping: it could overwrite a human's comment, and nothing cleared it when a review died. Re-filed with both as hard criteria.
• Completed stack shows the exact commit range reviewed; incremental runs show the delta range. DONE in feat(grug): every review says which commits it read #817 - and as a base..head range, not a single commit.
• Run-config disclosure section present and truthful. NOT DONE HERE - moved to the dashboard per operator decision 2026-08-02: useful to whoever runs Grug, not to whoever wrote the PR.
• No extra GitHub API writes beyond the existing upsert path. Satisfied - feat(grug): every review says which commits it read #817 adds no write path; it only changes what an existing write says.

Out of scope

• Resolved/Dismissed finding states - that is [grug] Living review: edit findings in place with Resolved/Dismissed/Addressed-in-sha state #556.
• Commit-range scoping of the review itself - [grug] Incremental per-push re-review: commit-range scoping, auto-pause, resume commands #557 owns that; this issue only SURFACES it.

Size: M

Refs #556 (living review states) and #557 (commit-range machinery). Part of #617 (structured/actionable comments).

Refs #791

---

Resolution 2026-08-02: split three ways - shipped (#817), re-filed (#819), and moved to the dashboard. Boxes above are ticked to record that each item has a DECIDED outcome, not to claim all three were built here; the two that were not are named as such on their own lines.

[coderabbitai]

🔗 Related PRs

#585 - feat(elder): make elder review durable and deep (waits for stable snapshot, merges independent findings, trusted-reaction learning) [merged]
#643 - feat(elder): Apex - Swift Hunt, in_progress checks, Markings Board [merged]
#647 - feat(elder): tiered review default - single coder + selective deep [merged]
#650 - feat(elder): async deep append after Tier-1 check (#646) [merged]
#658 - feat(elder): Markings v2 + review stack (CR-level comment UX) [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[quadseven]

Items 1 and 2 are implemented (38312bc). Item 3 is not, and I think it should be dropped - it reverses a decision this same epic already made.

Item 3 asks to "disclose the run posture in a collapsed section: tier used (single-arm/deep), settle window, arms, model labels". That content used to exist and was deliberately cut under #802, and the reasoning is still in the code at dispatch.py:1360:

Phase, Status and Check-run used to live in the fold. All three were cut: Status restated the summary line verbatim, Phase ("Tier-1 coder arm") is internal vocabulary that means nothing to the author, and the check-run name is already a row in the PR's own check list.

Both this issue and #802 are children of #791, so the epic currently contains an instruction to add back what it removed. Implementing item 3 as written would silently reverse that without anyone deciding to.

Worth separating the two motivations behind it:

• "Is this result current?" - that was the real gap, and item 1 fixes it. The board now says it is re-reading rather than showing a stale verdict.
• "What configuration produced this?" - this is the part fix: two email defects found by watching grug review its own PR #802 judged to be noise for the PR author. It is genuinely useful, but to an operator debugging grug, not to someone reading their own PR. That audience already has the check-run and DD.

Recommendation: close item 3 as superseded by #802, or re-file it as an operator-facing surface (a DD facet, or a reply to /grug why) rather than a section on the author's board. Leaving this issue open on item 3 alone is what would keep #791 from closing.

Items 1 and 2 shipped:

• Re-review flips the board to "Grug read new markings now. Wait." and names which push it is reading. Edit-only (create_if_absent=False), so it costs no email.
• Looked at: now discloses scope on a full base..head run too, not just a Living Hunt delta.

(Edited: an earlier version of this comment lost a phrase to shell backtick substitution.)

[quadseven]

Resolved by operator decision, 2026-08-02. Closing this in favour of three successors so the epic is not blocked on one ticket carrying three unrelated verdicts.

Item 1 - in-progress note. Implemented, reviewed, removed before shipping. Re-filed as #819 with the two failure modes that killed it as hard acceptance criteria: it must pass app_id (it could otherwise overwrite a human's comment), and every terminal path must clear it (a crashed review would otherwise leave "Wait" up forever with the findings already erased).

Item 2 - commit-range disclosure. DONE, shipped in #817. Every run now discloses what it read, and as a base..head range rather than a single commit. Previously only Living Hunt delta runs said anything, so a full review - the common path, and every first review of a PR - disclosed nothing.

Item 3 - run posture (tier, settle window, arms, model labels). Moves to the dashboard, not the author's PR comment. Operator call: it is useful to whoever runs Grug, not to whoever wrote the PR. Filing separately against the web surface.

Correction to my earlier comment on this issue: I claimed implementing item 3 would reverse #802. That was overstated. #802 cut "Phase, Status and Check-run" - only tier overlaps item 3's four sub-items, and the visible body still carries tier disclosure ("Fast look only - deep look may add more"). It was a reasonable product call, but it was a spec amendment, not a redundancy, and I should have said so.

[github-actions]

🚧 Reopened by the close-completeness guard — closed as completed but still has 4 unchecked item(s):

• Pushing to an open PR flips the stack comment to an in-progress note within the enqueue path (edit-i
• Completed stack shows the exact commit range reviewed; incremental runs show the delta range.
• Run-config disclosure section present and truthful (from the actual dispatch parameters, never hardc
• No extra GitHub API writes beyond the existing upsert path (edits, not new comments).

Finish + check them, or close it as "not planned" / add a force-close label. Exempt sections (Out of scope / Blocked by / Reverse-if-wrong / Candidate / Further notes) aren't counted.

[quadseven]

Re-closing after amending the acceptance criteria to reflect the three-way split. The guard was right to reopen: two of the four boxes described work that did NOT happen here, and ticking them as-written would have claimed otherwise. Each line now names its actual outcome - shipped in #817, re-filed as #819, or moved to the dashboard.