[foxnne]

Two process rules and a check for the second.

• CONTRIBUTING.md: auto-merge is turned on only after review is done and the maintainer said yes, by the maintainer or whoever merges on their yes, never by the session that opened the PR (app: a demo sets the person's panes aside #623/app: the history grip fills toward the history's cap #624 merged before review; follow-ups app: a demo's hold on the panes survives a failed copy and a launch #639/app: the history grip counts what the history keeps #640).
• docs/AUTOMATION.md: a press or release lands on an anchor a widget publishes (region:<name>, fizzy.*), never on the window or an offset from an anchor (layout: a place's arrange button grows out of its corner #634).
• sdk/tape/Tape.zig: Tape.firstButtonOffAnchor finds a press or release whose aim (the last move or trace since the last keyframe) has an empty tag or a dx/dy nudge. Unit-tested in the tape library.
• tests/demos.zig: every bundled demo is checked with it as it plays.
• tests/integration.zig: the soak verdict test asserts soak presses only on anchors (firstButtonOffAnchor() == null), now that layout: a place's arrange button grows out of its corner #634 aims it at region.arrange:Sidebar.
• ViewDrag.drawZones marks each bubble of a place's drop wheel as an anchor while a view is carried over it (drop:center/left/right/top/bottom/remove), and Script.drag takes a via target, since a drop's anchor only appears once the drag has moved. The tour now drops on drop:right via region:Main instead of an offset.

Review (PM reviewer pass)

code-review at medium. Fixed: the soak-only test exempted its only tape and so checked nothing; trace did not update the aim; a keyframe did not reset it; a drop (release) was unchecked; the doc named region.* and region.arrange:<name> as if they existed on main and said soak was already fixed; the paragraph above described offsets the rule now forbids for buttons; the check lived in a test file rather than with the tape; rule 5 did not say who turns auto-merge on. Declined: forbidding non-centre fractions (x, y are relative to the anchor's rect, not a coordinate), and enforcing the rule in Tape.parse (it would refuse live tapes and an agent's strokes; firstButtonOffAnchor is the seam a parse-time check would call).

Local gates on macOS (this reviewer): scripts/check-fmt.sh, zig build test (117/117 steps, includes the new Tape test), zig build test-integration (81/81 steps, every bundled demo passes the check), zig build check-tests-reached. No fizzy launched.

The original author's cloud session had no zig; its CI run was green on the first head.

🤖 Generated with Claude Code

[foxnne]

Integration tests (Linux) fail on 388054b: demos.test.demo: every bundled demo plays… returns error.ButtonOffAnchor with demo 'tour': op 13 presses or releases off an anchor. The other 627 tests pass.

This is the stricter Tape.firstButtonOffAnchor from 388054b catching the bundled tour (it counts release and trace, and a keyframe clears the aim). My original check passed the tour. The first s.click calls in src/editor/demos/tour.zig all aim at a bare tag, so I can't tell by reading which op 13 is. It needs a local run that prints the op's index and target.

I have no zig in this session, so I can't reproduce or push a fix. Auto-merge is on, so this red check holds the merge. Either the tour's op 13 gets re-aimed at an anchor, or the tour is exempted with a reason.

---

Generated by Claude Code