| name | ck-feature-frontier |
| description | Use when choosing what Ck framework work to build, harden, optimize, or prioritize next; not for live bug triage, campaign execution, or performance measurement. |
ck-feature-frontier — where the framework could advance next
Overview
This is the vetted portfolio of candidate framework advancements for CkFoundation /
CkGameplayDebugger / CkTests, grounded in what the code actually contains as of 2026-07-02.
Everything here is a candidate — evidence-backed and scoped, but not committed roadmap; the
maintainer picks. Every "Gap" and "asset" claim below was re-verified against the repos on the
campaign date (re-verification commands in the final section).
The ruling (maintainer, 2026-07-02 — binding)
"Advancing the framework" = ROBUSTNESS and PERFORMANCE first; debug tooling next (tooling
usually lands in CkGameplayDebugger — debug tooling is explicitly valued). Stale/outdated
branches MUST NOT be touched — they are context, not work items.
In-repo record: .claude/reports/DECISIONS.md §25. Consequences:
- Rank candidates robustness > performance > tooling. Within a class, prefer the candidate whose
first step is a measurement or a failing test.
- A proposal that amounts to reviving a fenced branch (below) is out of bounds — even if
re-described in new words.
The fence — do-not-resurrect branches
The canonical registry lives in ck-failure-archaeology (§"Stalled-branch registry —
DO-NOT-RESURRECT") — per-branch tip dates, history, and assessments live there, once. This section
only states the rule and the headline names (verified on origin 2026-07-02):
feature/dependency-injection-entity-script · feature/entity-script-as-script-struct ·
upgrade/ue5.7 · dev-bb-5.7 · dev-bb · feature/editor-time-construction-scripts ·
feature/message-without-name · bugfix/test/investigate-cheat-in-build ·
feature/entity-replication-channel · feature/flow-graph-module · feature/ability-traits ·
feature/modular-traits · shelf/dummy-changes-to-debug-networking-asserts · plus the backup
fossils backup/before-reverting-f8a3a55 and backup/pre-exporter-rewrite-2026-05-28 (insurance
snapshots, not work). Do not check them out, merge them, or restart their feature under a new name
without the maintainer's explicit say-so.
feature/registry-handle-storage-support is cited by a live comment
(Source/CkEcs/Public/CkEcs/Registry/CkRegistry.h:289) but does not exist on origin — don't
go hunting for it. Branches with tips from ~June 2026 onward are likely active sibling work
(e.g. the iskm/save-load lines) — coordinate, never adopt.
When NOT to use this skill
| You actually have… | Load instead |
|---|
| A live bug / build break / crash to fix | ck-debugging-playbook |
| The entity-teardown / signal-unbind defect campaign | ck-lifecycle-teardown-campaign |
| "Has this been tried before?" | ck-failure-archaeology |
| How to benchmark / make a perf claim | ck-performance-and-analysis |
| Process, phase-gates, multi-session discipline | ck-methodology |
| How to add a debugger view once a tooling candidate is picked | ck-gameplaydebugger-extension |
Reference files — load only what the task needs
Section numbers cited elsewhere in this skill point into these files.
| Topic | Read |
|---|
| The portfolio | references/portfolio.md |
Prioritization
- Robustness first (ruling): candidates 1, 2, 3 — and 6, whose payoff is robustness even
though it ships as tooling.
- Performance second: 4 and 5. Both are deliberately measure-first — the FIRST step is the
benchmark, and "measured: negligible" closes them honorably. No perf change lands without the
A/B number (root non-negotiable #7).
- Tooling third: 7 and 8 — explicitly valued by the maintainer, and both land in their proper
homes (7 in CkGameplayDebugger; 8 in CkFoundation's own analyzer module).
- Within a class, prefer the candidate whose first step is a measurement or a failing test — all
eight above were shaped to satisfy that.
Common mistakes
- Reviving a fenced branch by paraphrase. "Dependency injection for entity scripts" in new
words is still
feature/dependency-injection-entity-script. The live DI-flavored mechanisms are
CkProvider + ContextOwner (root CLAUDE.md lingo table); anything beyond them needs the
maintainer.
- Building before measuring on candidates 4/5. The define flip and the group prototype are
cheap to write — that is exactly why the ruling is benchmark-first; an unmeasured "optimization"
is a review rejection.
- Treating
in_place_delete as a work item. It is settled design (DECISIONS.md §45 — the
ungating was deliberate); the open part is only candidate 5's measurement. "Restoring the debug
gate" is not a cleanup — it would revert a considered 2026 storage decision.
- Landing debug tooling in CkFoundation runtime modules. Overlay/inspector work belongs in
CkGameplayDebugger (
ck-gameplaydebugger-extension); only the analyzer (already a CkFoundation
UncookedOnly module) and the AS checker (generator-adjacent) live here.
- Presenting this portfolio as a roadmap. Every entry is a candidate. When you pitch one,
carry its evidence (the file:lines above), not its conclusion.
- Skipping the ask. If developing a candidate surfaces a genuine convention fork, add it to
.claude/reports/ADJUDICATIONS.md instead of inventing policy (root non-negotiable #6).
Provenance and maintenance
Authored 2026-07-02 (documentation campaign; CkFoundation HEAD 7330c1bab). All counts and lines
below drift — re-verify before relying on them. Commands are Git Bash from
d:\Repos\BusterBlock\Plugins\CkFoundation (the Grep/Glob tools are blind under Script/ — use
rg --no-ignore):
- Ruling + fence records:
.claude/reports/DECISIONS.md §25 + §45 (ADJUDICATIONS.md A3 is a
resolved tombstone pointing at §45).
- Branch list:
git for-each-ref --sort=-committerdate --format='%(refname:short) | %(committerdate:short)' refs/remotes/origin
- Teardown TODOs:
rg --no-ignore -n "teardown is mid interaction" Source/CkInteraction (expect 2)
· rg --no-ignore -n "bullet-proof way" Source/CkRelationship (expect 3)
- GC census:
rg --no-ignore -c "TStrongObjectPtr<" Source/ --glob '*_Fragment.h' (13 total) ·
same for TWeakObjectPtr< (35) and TObjectPtr< (4) · tripwire:
rg --no-ignore -n "IsInGameThread" Source/CkCore/Public/CkCore/IO/CkDeferredAssetInit_AngelScript.cpp (:521)
- Snapshot:
rg --no-ignore -c "CK_REGISTER_SNAPSHOTABLE\(" Source/ --glob '*.cpp' (sum = 119) ·
rg --no-ignore -c "struct (CK\w+_API )?FFragment_" Source/ --glob '*_Fragment.h' (sum = 295) ·
policy doc at Source/CkEcs/Public/CkEcs/Snapshot/CkSnapshot_Policy.h:9-19
- Net-param copy:
rg --no-ignore -n "CK_DISABLE_NET_PARAM_COPY_PER_ENTITY" Source/ (6 Build.cs
sites all =0 + 1 consumer at CkEntityLifetime_Utils.cpp:478)
- Groups / deletion policy:
rg --no-ignore -n "static_assert" Source/CkThirdParty/Public/CkThirdParty/entt-3.16.0/src/entt/entity/group.hpp
(in-place-delete rejection at :697) · funnel at Source/CkEcs/Public/CkEcs/Registry/CkRegistry.h:162 ·
global trait at Source/CkEcs/Public/CkEcs/Handle/CkHandle.h:71-77 — confirm it is still
ungated via grep -n "#if" Source/CkEcs/Public/CkEcs/Handle/CkHandle.h | head -1 (first hit :79,
i.e. after the trait) · origin chain: git log --format='%h %ad %s' --date=short -G component_traits -- Source/CkEcs/Public/CkEcs/Handle/CkHandle.h
(returns 6b54d2e38 relocate + 745507381 introduce-gated) and
git log --format='%h %ad %s' --date=short -G CK_DISABLE_ECS_HANDLE_DEBUGGING -- Source/CkEcs/Public/CkEcs/Handle/CkHandle.h
(returns 06938bba3 = the deliberate ungating, + 232ae5e06 param rename). -S cannot find the
move/ungate commits — the string count never changed.
- AS surface:
ls Script/Generated/*.as | wc -l (274) · emitted-namespace claim at
Script/CLAUDE.md:113 · commandlet scope in
