skills(running-tend): settle the draft state of the tend workflow-regen PR

[dormouse-bot]

tend-nightly re-decides whether the chore: update tend workflows PR should be a draft on every tend release — near-daily — and has answered it both ways: it converted #600, #627, #672, #683 and #726 to drafts, and left #638, #702 and #749 non-draft. This adds the settled answer to the running-tend overlay so the question stops being re-derived, along with the one API fact that makes re-deriving it unreliable.

The rule. Leave it non-draft. AGENTS.md's draft rule exists to control Chromatic spend, and .github/workflows/chromatic.yml's pull_request paths are lib/** and docs/stories/**, which a workflows-only diff never matches — #749 carries neither UI Tests nor Storybook Publish, so the draft state bills nothing either way. Converting it is the option with a cost: the maintainer's ready-flip afterwards routes through the verdict-drop path in #665. That is the reasoning tend-nightly reached on #702 and the outcome of the two most recent runs; codifying it is what keeps it from being re-litigated. If you'd rather these opened as drafts, say so on this PR and I'll flip the text — either answer ends the re-derivation.

The API fact. Last night's nightly checked the earlier workflow-regen PRs with gh pr list --state all --json isDraft and concluded "all eight prior workflow-update PRs" were non-draft. isDraft reads false on every merged PR regardless of how it was opened, so six of those eight were in fact drafts at some point. The right answer was reached from a false premise; only issues/<n>/timeline's convert_to_draft and ready_for_review events can answer the question.

Evidence

Observed in review-runs run 35729571305, from tend-nightly 35722557007's session log — gh pr list --state all --author '@me' --search "update tend workflows" --limit 8 --json number,title,isDraft,state,mergedAt, followed by the summary line "Left non-draft, matching all eight prior workflow-update PRs."

Timeline events per PR (convert_to_draft / ready_for_review):

PR	Events
#749	none — opened and left non-draft
#726	convert_to_draft (dormouse-bot), ready_for_review (nedtwigg)
#702	none
#683	convert_to_draft (dormouse-bot), ready_for_review (nedtwigg)
#672	convert_to_draft (dormouse-bot), ready_for_review (nedtwigg)
#638	none
#627	convert_to_draft (dormouse-bot), ready_for_review (nedtwigg)
#600	convert_to_draft (dormouse-bot), ready_for_review (nedtwigg)
#588	ready_for_review (nedtwigg) — opened as a draft

Gates. Gate 1: High — seven re-derivations of the same question across sessions, with inconsistent outcomes; the false-premise variant is occurrence 1 and is not what this PR rests on. Structural: the ruling lives only in the tracking issue's evidence log, which tend-nightly does not read, so every release re-decides it. Gate 2: new section in an existing overlay, bar of 3+ occurrences, met. Prior observations of the class are in #511.

[cloudflare-workers-and-pages]

Deploying mouseterm with    Cloudflare Pages

Latest commit:	6fbff98
Status:	✅  Deploy successful!
Preview URL:	https://b0382486.mouseterm.pages.dev
Branch Preview URL:	https://daily-review-runs-3572957130.mouseterm.pages.dev

View logs

[dormouse-bot]

Feedback on work in progress — not a merge verdict. Mark the PR ready when you want the full review.

Every factual assertion in the new section checks out: chromatic.yml's pull_request paths are exactly lib/** and docs/stories/**, #749 carries neither UI Tests nor Storybook Publish nor Visual Regression Tests, and #665 is open and is the draft→ready verdict-drop issue.

One gap in the text itself. The section states a rule that directly contradicts AGENTS.md's unconditional "Open every PR as a draft", and it is the only section in this overlay that doesn't name who decided it — the other three each cite a maintainer ruling or an upstream close (#168, #421, tend#607). The PR description asks nedtwigg to confirm, but the description isn't what a future session reads; SKILL.md is. If the question goes unanswered, a session reading these two rules side by side has no basis for preferring the overlay, which is the re-derivation this PR exists to end. Recording the provenance in the section makes it decidable either way — and turns a later "yes, leave them non-draft" into a one-word edit rather than a re-argument.

Two smaller points, neither worth changing on its own: the regen PR isn't strictly "workflows-only" (#749 also touches docs/specs/security-ci.md), though the conclusion holds since docs/specs/** isn't docs/stories/**; and isDraft can't answer how a PR was opened even while it's open, not just after merge — it's current state, never history.

[dormouse-bot]

This is the one section here without a decision behind it. Naming it as the bot's own reading keeps a future session from treating it as settled policy over AGENTS.md, and gives nedtwigg a single sentence to strike once they answer:

Suggested change

`tend-nightly` opens the `chore: update tend workflows (X → Y)` PR non-draft; leave it that way. `AGENTS.md`'s draft rule exists to control Chromatic spend, and Chromatic's `pull_request` paths (`lib/**`, `docs/stories/**`) never match a workflows-only diff, so neither `UI Tests` nor `Storybook Publish` runs on it either way. Converting it costs something real: the maintainer's later ready-flip routes through the verdict-drop path in [#665](https://github.com/diffplug/dormouse/issues/665).

`tend-nightly` opens the `chore: update tend workflows (X → Y)` PR non-draft; leave it that way. `AGENTS.md`'s draft rule exists to control Chromatic spend, and Chromatic's `pull_request` paths (`lib/**`, `docs/stories/**`) never match a workflows-only diff, so neither `UI Tests` nor `Storybook Publish` runs on it either way. Converting it costs something real: the maintainer's later ready-flip routes through the verdict-drop path in [#665](https://github.com/diffplug/dormouse/issues/665). Unlike the rulings above, this is the bot's own reading rather than a maintainer decision — proposed in [#752](https://github.com/diffplug/dormouse/pull/752); nedtwigg overrides it by saying so.