Ledgework: Mantle, Ledge Climb and Wall Climb for Any Character
- Author
- claude-code @ auraAgent
- Opened
- Nov 17, 2026
- From
- contrib-mzbsee1a
- Into
- main
- Merged
- Nov 17, 2026
From High Ground's Ridge Run at normal speed (keyboard, 2026-11-17): after the ledge jump, a run-up with forward held caught the roof eave of a house as a hang and hung there for 25 s with forward still held. Cause: hangUpdate only pulls up on an armed push (pushArmed, set after easing the stick below u 0.45 for 0.12 s), so a push carried over from before the catch never arms. A replay starting from no input pulled up in 0.9 s. Changes (ClimbMover.component/init.luau):
- heldUpFor / HELD_PUSH_TIME = 0.6: a push held toward the wall (u > 0.6) for 0.6 s also arms the pull-up. A catch still holds for a moment and does not flick straight into a pull-up. heldUpFor resets in tryHang with pushArmed.
- hangNudgeFor: a push held partly toward the wall (0.3 < u <= 0.6) that shimmies nowhere this step (shimmied false, set from the shimmy and corner paths) pulls up after 0.4 s if findLedge finds a top (same options as the pull-up), or climbs over a lip. Pushes of u <= 0.3 (sideways) and shimmies that move are unchanged. Reset in handBack. Checked in High Ground: full Ridge Run on keyboard at normal speed finished (all checkpoints, no falls) after the change; before it, 2 of 3 runs hung at that eave. Plain-world set and ledge checks rerun before release.
Source world: 8991520f-742b-4e57-b532-81c35cd08ded Contributed paths:
- climbKit.package/ClimbMover.component/init.luau
- climbKit.package/README.md
- climbKit.package/package.yaml
- climbKit.package
- climbKit.package/ClimbMover.component
- scenes/main.scene/authoring/entities/page-03.json
- scenes/main.scene/authoring/state.json
- scenes/main.scene/build.luau
- scenes/main.scene/scene.json
6 files changed
- ChangedclimbKit.packageFolderBeforeAfter
- ChangedclimbKit.package/ClimbMover.componentFolderBeforeAfter
- ChangedclimbKit.package/ClimbMover.component/README.md7.7 KB → 7.8 KBBeforeAfter
- ChangedclimbKit.package/ClimbMover.component/init.luau67.0 KB → 74.0 KBBeforeAfter
- ChangedclimbKit.package/README.md16.4 KB → 18.5 KBBeforeAfter
- ChangedclimbKit.package/package.yaml359 B → 359 BBeforeAfter
Checks
No checks have been posted.
Reviews
No reviews yet.
Conversation
- claude-code @ auraAgentNov 17, 2026
Review: request changes (small). Reviewed the installed copy in High Ground (climbKit.package, ClimbMover.component/init.luau), including the hang unshadow block added after the PR was opened. I could not read the PR base (Ledgework main) from here, so 'no other behaviour changed' is checked against the PR description and the touched regions only.
OK:
- heldUpFor / HELD_PUSH_TIME declared at init.luau:165-166, before tryHang (526) and hangUpdate (1282); heldUpFor reset with pushArmed=false and easedFor in tryHang (init.luau:587-589). Arming rule init.luau:1340-1342: a push carried over from the catch needs 0.6 s of u > 0.6 (one dropped-input frame resets it), so a catch still does not flick into a pull-up; a held push now pulls up after 0.6 s. Correct.
- hangNudgeFor declared at init.luau:138 next to pinnedFor, reset in handBack (init.luau:443) and in the else branch whenever the condition breaks. shimmied is false by default, set from moved and true on tryCorner. Nudge rule only for 0.3 < u <= 0.6 and not shimmied, runs after the drop return (u < -0.6 never reaches it), uses the same ledgeOpts/minHeight/probeDistance as the pull-up and falls back to climbOverLip. Sideways (u <= 0.3) and moving shimmies are untouched. Note: a shimmy blocked at a dead end (no move, no corner) with u in (0.3, 0.6] does count toward the nudge, which is intended per the description.
- Unshadow block: unshadowAt declared at init.luau:140 before use; clock (87), setFeet (208) and Physics.raycast use match tryHang. It moves only along +wallN (y unchanged), only when overHead(p) is true, and only to a spot that is clear and passes Ledge.fits, through setFeet (MAX_STEP refusal). A normal hang with a clear ray is not moved.
Changes requested:
- init.luau:1330 - overHead returns hit ~= nil, but the catch's own check (init.luau:574-575) ignores over.startedInside and excludes the tryHang exclude list. As written a hang whose head ray starts inside a collider (accepted by tryHang) is stepped out from the wall every 0.2 s even though nothing is over it. Use 'hit ~= nil and not hit.startedInside' so the per-frame check agrees with the catch.
- init.luau:1334 - steps go to 0.6 m, while a fresh catch only steps out < 0.44 m (init.luau:559). The comment says 'as a fresh catch would'; either cap at the same 0.44 or say why the hang may sit further out (hands visibly off the lip at 0.6 m).
- Versioning: the unshadow change is not in the changelog. package.yaml:2 still says 1.1.24 and README.md has no 1.1.25 entry (last is README.md:146, 1.1.24, which does not mention it). Bump package.yaml to 1.1.25 and add a README 1.1.25 line for the unshadow step before pushing to this PR.
The 1.1.24 part (held push, nudge) is good to go as is; README 1.1.24 entry and package.yaml 1.1.24 agree with it.
- claude-code @ auraAgent
Review: request changes (small). Re-reviewed the installed copy in High Ground (climbKit.package, ClimbMover.component/init.luau). PR base still not readable from here; checked against the first review and the touched regions.
Requested changes from the first review - done:
- init.luau:1333 overHead now returns 'hit ~= nil and not hit.startedInside', the same reading as tryHang's clearOver (init.luau:567).
- Step-outs agree: tryHang (init.luau:572-578) and the hang-time block (init.luau:1335-1345) both try 0.04 m steps out along the wall normal to 0.6 m (15 positions each; float sums checked, both stop at 0.5999...), both need Ledge.fits(.., 0.02, ..) plus a clear head ray to hangTopY/topY + 0.4, and rise <= 0 counts as clear in both. The README 1.1.25 entry explains why the cap is 0.6 now.
- package.yaml:2 is 1.1.25; README.md:148 has the 1.1.25 entry covering the catch and hang-time step-outs.
1.1.24 parts unchanged since the first review: heldUpFor/HELD_PUSH_TIME (init.luau:167-168, reset 592, arming 1350-1351), hangNudgeFor (138, handBack reset 445, rule 1415-1435), shimmied (1376, 1393, 1397). Declaration order OK: clock 87, MAX_STEP 109, hangNudgeFor 138, unshadowAt 140, heldUpFor 167 all before tryHang (526) and hangUpdate; exOver/clearOver/goodSpot (560-571) and overHead (1327) are declared before use.
Changes requested:
- init.luau:553-554 - maxShift is checked on px/pz BEFORE the step-out; the final spot can be up to 0.6 m further and only MAX_STEP (582) runs after it. The comment says 'never snap further than the probe could reach'. Check maxShift against spot (after the loop, beside the MAX_STEP test) or widen it by 0.6 on purpose and say so.
- init.luau:557-559 and README.md:148 - 'A roof seen from under its eave stays refused: its edge is further out than a hang reaches' is not enforced by code, it is an assumption about the geometry. Before, a blocked head ray refused the catch; now it steps out instead, so an eave whose soffit ends within about r + wallGap + 0.6 m of the wall (~0.95 m at radius 0.3) that findLedge accepts (nothing over the edge, Ledge 251) is now caught in front of the eave, where before it was refused. That hang is clear overhead and can pull up, so it may be fine (it is the same geometry as the balcony slab this targets), but the comment and changelog should say that (eaves shallower than ~0.6 m plus radius are caught out in front) rather than claim they stay refused - or add a real test (e.g. refuse when the overhead hit is sloped, normal.y < 0.9) if they must stay refused.
Nits (non-blocking): the hang-time overHead/fits (init.luau:1331, 1339) exclude only the entity, not the exclude list a public:tryHang(..., exclude) caller passed (exOver, init.luau:560-561), so a hang caught past an excluded collider can be stepped out again by it; keep the exclude list for the hang if that API case matters. PR title and description still say 1.1.24 and describe only the held push; update them to 1.1.25 with the step-out, and make sure the 1.1.25 commit is on contrib-mzbsee1a (head shown is ca564ac9).
- claude-code @ auraAgent
Review: approve. Re-reviewed the installed copy in High Ground (climbKit.package 1.1.27: ClimbMover.component/init.luau, ClimbMover.component/README.md, README.md, package.yaml). PR base not readable from here; checked against the second review and the touched regions.
Second review's requests - done:
- init.luau:606-609 maxShift is now checked on the final spot after the step-out loop (598-604), beside the MAX_STEP test (611-612), and widened by 0.6 on purpose with a comment saying so (r + probeDistance + 0.6 + 0.6).
- init.luau:577-583 comment and README.md:150 (1.1.26) now say a shallow eave (underside within ~0.95 m = r + wallGap + 0.6) is caught out in front and pulled up onto, and a deeper one is refused because no spot within 0.6 m has a clear head. That matches the code (goodSpot/clearOver, 588-597; refusal 605).
1.1.27 stuck nets - OK:
- stuckFor/stuckAt/STUCK_TIME/STUCK_DIST declared at 138-141, before handBack (440), startClimb (511), tryHang (551), mantleUpdate (741), climbUpdate (1042), hangUpdate (1347).
- Climb rule 1221-1243 and hang rule 1500-1521 run last in each update, only for u > -0.5 and uu+ss > 0.36, anchor re-set whenever the body is more than 5 cm (squared compare) from stuckAt, fire at 2 s. Reset in handBack (461-462), startClimb (529-530), over-lip end (766-767), tryHang (623-624) and the else branch when the stick eases. At rest (magnitude <= 0.6) it never counts; climbing (1.6 m/s) and shimmying (1.1 m/s) leave 5 cm in a few frames; a mantle runs mantleUpdate, not these; a corner turn returns early from hangUpdate before the rule. Hang version is kickOff(cc, false, 'no way up', true): away=false (1 m/s off the wall, no up), plus lockReleasedLedge, i.e. a drop.
No-climb lock - OK: noClimb* (145-148) set only under reason == 'no way up' in kickOff (791-796); startClimb (514-518) refuses only dot > 0.8 within NO_CLIMB_DIST 2 m while clock < noClimbUntil (3 s). feet() (219) and the noClimb locals precede kickOff (789) and startClimb (511). The public tryClimb (1616-1617) answers public.state == 'climbing', so a refused start reports false; idleUpdate callers (678, 687) just return.
hangExclude - OK: declared 156, set from exOver in tryHang (585-587), used by the hang-time overhead ray (1394) and Ledge.fits (1402).
Docs - OK: package.yaml 1.1.27; README.md:146/148/150/152 entries for 1.1.24-1.1.27 agree with the code; ClimbMover.component/README.md:63 lists 'no way up' for onClimbEnd and onHangEnd.
Nits (non-blocking):
- By design the hang net also drops a player holding sideways at a ledge's dead end (no shimmy, no corner) for 2 s, which can be a long fall on a high ledge. The comment at init.luau:1497-1499 says 'or along it', so intended; mention it in README.md:152 so authors are not surprised.
- kickOff sets the 3 s no-climb lock for a hang 'no way up' too (isHang true); ClimbMover.component/README.md:63 only mentions the lock for onClimbEnd.
- PR title/description still say 1.1.24 and describe only the held push; update them to 1.1.27 before merge.
Sign in to comment or review.