菜单
Deep behavioral correctness review (pass 3): concurrency, algorithms, SIMD parity, pipeline determinism, resource lifetime, outliers, and test-coverage gaps. Use when the user asks for deep correctness, concurrency review, determinism audit, race hunting, or /zig-deep-correctness-review. Goes beyond idiom/surface reviews.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
ZeroLight-Framework (Zig 0.16, SDL3, SDL_GPU) specialist routing: design, implement, debug, PR/diff review, and multi-phase review passes. Use when the user asks for zig-design-specialist, zig-specialist, zig-debug-specialist, or zig-review-specialist; when zig build, zig test, or shader compile fails; when reviewing a PR or branch diff; when planning ECS, DataSystem, SimulationPipeline, pathfinding, collision, AI, or threading changes; or when assessing emergent-gameplay architecture readiness. Slash: /zig-workflows.
Assess ZeroLight-Framework architecture for scalable emergent gameplay simulation. Use when the user asks how ready the engine is for emergent gameplay, architecture assessment, extensibility review, or runs /architecture-assessment.
Multi-agent Zig review of the pathfinder module for coherency, cohesion, and standards adherence. Use when the user asks for pathfinder review, pathfinding module review, /pathfinder-review, or a module-wide pass on src/game/systems/pathfinding/.
Multi-agent Zig best-practices review of large hot subsystems: review, adversarial verify, then synthesize durable lint/agent/doc guidance. Use when the user asks for best-practices review, Zig idioms pass, durable lint guidance, or /zig-best-practices-review.
ZeroLight-Framework (Zig 0.16, SDL3, SDL_GPU) specialist workflows: design plans, implementation, debug/triage, PR/diff review, pathfinder module review, architecture assessment. Use when the user asks for zig-design-specialist, zig-specialist, zig-debug-specialist, or zig-review-specialist; when zig build, zig test, or shader compile fails; when reviewing a PR or branch diff; when planning ECS, DataSystem, SimulationPipeline, pathfinding, collision, AI, or threading changes; or when assessing emergent-gameplay architecture readiness.
| name | zig-deep-correctness-review |
| description | Deep behavioral correctness review (pass 3): concurrency, algorithms, SIMD parity, pipeline determinism, resource lifetime, outliers, and test-coverage gaps. Use when the user asks for deep correctness, concurrency review, determinism audit, race hunting, or /zig-deep-correctness-review. Goes beyond idiom/surface reviews. |
Pass 3. Do not re-report idiom/naming/stdlib/surface ReleaseFast issues already covered by best-practices or idiom-lint. Focus on behavioral correctness.
Agents: zig-review-specialist.
Themes: .grok/skills/zig-workflows/references/module-presets.md →
deep-correctness-review.
One zig-review-specialist per theme unit (background: true,
capability_mode: read-only). Use the preset focus text as the unit brief.
For each finding give concrete failure: inputs/state → wrong behavior / crash /
leak / nondeterminism. Cite real file:line. Prefer few high-confidence findings
over speculation. Clean units with zero findings are valid.
Kinds: bug, race, determinism, numerical, leak, test-gap, anomaly,
contract. Severity: high|medium|low. Also: title, detail, fix_direction,
durable, durable_mechanism, durable_rationale.
For each finding, spawn a verifier that defaults to skepticism:
CONFIRMED only with a concrete failing case or confirmed coverage gap.
Keep CONFIRMED/PLAUSIBLE with is_real.
Split cleanly:
If nothing survives verification, report that the reviewed surface is clean — that is a valuable result.
