notque/vexjoy-agent / skills/game/game-dev
game-dev
FlaggedGame development: design, asset pipelines, Phaser, Unity CLI and builds, vanilla JS frontend, GM mode, match booking.
references/quality-gates.md
View on GitHub# Quality gates
## Before implementation
- Target instructions and local GM skills loaded completely.
- Current code, live state, player evidence, and prior feedback inventoried.
- One active short ADR and version envelope accepted.
- DAG/spec hashes match and applicability validates.
- CPU harmony passes when applicable.
- Single writers and worktree boundaries are explicit.
## Focused implementation gates
- Behavior-changing work has contract tests or a characterized baseline.
- Backend and frontend use one canonical rule/grade/result source.
- Deterministic seed/order/replay and CPU-ignore paths remain valid.
- 1981/2026 or target era goldens cover historical boundaries when relevant.
- 360px mobile, keyboard, semantics, contrast, and reduced motion are checked.
- Player-facing copy passes the target anti-AI/no-em-dash doctrine.
- Analytics measure benefit and harm; vanity metrics alone never pass.
## Adversarial review gate
Run independent code/perspective, system/integration, and security/privacy
lenses. Converge duplicates and classify findings. Any critical issue blocks.
Important issues must be fixed or explicitly returned to the active short ADR.
## Predeploy game-design gate
Apply these game-design packets to the implemented player moments:
1. Core Loop Extractor: cue to next intention still closes.
2. Craft Critique: presentation supports the promised fantasy.
3. Attribution Audit: players can connect cause, result, and next choice.
4. Fogg Behavior Audit: motivation, ability, and prompt align without coercion.
Fix concrete misses before deployment. Record observed/documented/measured/
inferred evidence and the smallest loop test.
## Release authorization
A valid scoped owner or fast-development approval record can satisfy the human
approval question. It cannot skip branch, worktree, privacy, focused tests,
protected-repository, rollback, canonical deploy, exact-SHA, service, health,
asset, or route guards.
## Single deploy lease
- One integration owner holds one target-project deploy coordinator/lock for
staging and production.
- Immediately before acquisition, reread live SHA and every ready wave commit.
- Rebase or cherry-pick the accepted ready set into one clean integration
candidate; record included and explicitly deferred artifacts and commits.
- Validate every included artifact's complete prerequisite closure using its
declared exact path-set/content-tree hashes or full normalized source-diff
hash. A present top commit, matching subject, or clean cherry-pick is not
proof that its parent prerequisites landed.
- Recompute the aggregate candidate closure after conflict resolution and
immediately before staging. Reject missing paths, unexpected deletions,
undeclared overlap, or a closure hash that differs from the lease record.
- Broadcast the active owner, live base, candidate SHA, environments, and lease
expiry to every active lane.
- Reject a second staging/production attempt, stale base, changed candidate,
unlisted ready artifact, incomplete prerequisite/file-set/diff closure,
missing lock proof, or non-owner deploy.
- Keep the lease through exact-live verification or rollback, then release it
with the corresponding receipt.
## Staging and production proof
- Use only the target repository's supported deploy path.
- Require the valid active deploy-lease record before staging or production.
- Staging serves the intended SHA and passes health/assets/routes/contracts.
- Production deploys the exact staged candidate unless the repository contract
explicitly rebuilds and proves equivalence.
- Verify release marker, service health, public assets, public/auth-gated routes,
timing or response contracts, and rollback pointer.
- On failure, stop or rollback through the canonical path; never hand-edit live
state without mirroring through the repository's emergency doctrine.
## Completion gate
Record exact live SHA, checks, review verdicts, release evidence, feedback
baseline/window, unresolved items, and next-wave dependencies. A missing
measurement remains `missing evidence`, never `done`.
## Conditional defect and evidence gate
- A reported defect enters implementation only with an exact-state
reproduction record: release SHA, state-class fingerprint, environment,
deterministic steps, expected/observed result, severity, privacy-safe
evidence hash, and independent confirmation.
- `repair_now` is allowed only for reproduced state loss, blocked core action,
duplicate write, sustained 5xx spike, truth-boundary breach, or accessibility
blocker. The active incident owner, rollback pointer, bounded repair scope,
and stop condition are mandatory.
- An actively observed production emergency may be contained before full
reproduction only when containment is narrower than continued harm. It may
not authorize feature work, and the incident record must be completed during
containment.
- Human and volume-dependent rows remain `evidence_blocked` until their exact
threshold is met. Zero observations are no signal. Qualitative rows require
the declared unprompted protocol; analytics rows require the declared
eligible denominator and privacy/small-cell guard.
- While blocked, the only allowed work is privacy-safe measurement,
recruitment, protocol execution, or reproduction. UI, simulation, economy,
copy, and progression changes are refused as speculative.
- Define each threshold once in a canonical evidence-gate registry. Every
disposition cites its row ID plus registry hash. The bounded consumer
manifest contains only consumer identity, artifact hash, row IDs, and the
registry hash; its closed schema rejects copied threshold fields. Narrative
reports link to a validated consumer receipt and are not parsed as prose.
## Canonical completion snapshot
- Preserve release history as evidence, but keep exactly one current snapshot.
- The snapshot records scope hash, exact release SHA, exhaustive status counts,
row-disposition hash, evidence-registry hash, superseded snapshot hashes, and
the verifier receipt.
- Counts must equal the declared scope cardinality. Each row appears exactly
once. Evidence-gated questions retain their future trigger but cannot appear
as valid unfinished implementation.
- A new release atomically replaces the current snapshot and links its
predecessor. Appended prose, document position, or a heading such as “final”
cannot override the machine-readable current snapshot.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |