[saworbit]

Summary

• [Bake] A bake applies the pending cuts with no undo step #1353: HFBakeSystem._bake_impl() calls root.apply_pending_cuts() and no caller recorded that move; the bridge's hf_bake bumped the revision by hand and ended the agent's undo chain. The step belongs to the callers, since the bake system has no undo manager: HFDockManageHandler.apply_pending_cuts_as_a_step() commits "Apply Pending Cuts" through commit_if_changed() (a step only when something was pending) before every dock bake that applies cuts: Bake, Bake Changed, the three Quick Plays (before the spawn's temporary move, so the capture never holds the spawn at the camera), Export Playtest, the game scene export, Validate Spawn and Spawn Preview. The bake then finds nothing to apply. hf_bake records MCP: apply pending cuts through the dispatcher's own capture and commit, so the revision moves through version_changed and hf_undo takes the cuts back; hf_bake itself is still no step. Both pending_cut KNOWN cases come off the undo sweep.
• [Cuts] Redo of Commit Cuts gives the committed cutter a name the commit did not #1354: not the commit. apply_pending_cuts reparents with add_child(child), and a node landing beside a same-named sibling is named @Box_0@N (both the box and the cut are Box_0); the record leaves @ names out (Every undo moves the brushes it touched to the end of the level, so a save after an undo rewrites the file #660) and the rebuild names it Box_0. The four container moves use add_child(node, true), Godot's readable-name rule, which numbers past the sibling as a new brush is numbered and keeps a mapper's stem. KNOWN case off.
• [Cuts] A caller of commit_cuts() never resumes when the level is freed mid-bake #1357: prepare_commit_cuts() and commit_cuts() return bake_system.answer_to(...), the [Bake] Let a caller of a level's bake resume when the level is freed mid-bake #1294 shape, and the brush system's prepare_commit_cuts returns false when the root is gone after its bake instead of touching it.

Docs: hammerforge_mcp/README.md, catalogue.py's hf_bake description, one SPEC sentence and the bridge design spec.

Before / After Behavior

• Before: Bake Now with a pending cut applied it for good; Quick Play's redo brought back its spawn but not the cuts; the committed cutter's record changed name across undo and redo; a caller of commit_cuts() on a closed tab never resumed.
• After: one undo after a bake puts the cuts back to pending, on the dock and over the bridge; a moved cutter keeps a readable name on every path; commit_cuts() callers get false when the level goes.

Related Issue

Fixes #1353
Fixes #1354
Fixes #1357

Checks

• test_undo_sweep 6/6 with no KNOWN left, test_bake_applies_cuts_as_a_step 3/3, test_commit_cuts_names_the_cutter 5/5, test_bake_survives 7/7, test_mcp_dispatcher_bake 24/24, test_cutting 34/34, test_quick_play 32/32, test_dock_history_and_playtest 8/8, plus the builder's wider set (test_bake 19 scripts, test_mcp 19 scripts, cuts, cordons, dock, prefab, registry)
• Each test file run before its fix fails with the expected messages (@Node3D@19 names, null answers, no step, undo_conflict)
• pytest hammerforge_mcp 110 passed; ruff clean; CI's gdformat and gdlint commands pass; check_script_warnings.py clean; .durations.json entry for the new bake test
• A fragment in changelog.d/, no edit to CHANGELOG.md