菜单
Drive the architect→workhorse flow — turn an approved plan into delegated Sonnet implementation with test and review gates, keeping the main (Opus/Fable) thread as coordinator.
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
Work a GitHub issue end-to-end — fetch it, verify the claim, turn it into an architect spec, and drive the delegated implementation flow with correct issue references in branches and commits.
Checklist for adding support for a new GGUF model architecture (or a variant of an existing one) to SharpInference.
Cross-check SharpInference numerics against llama.cpp and run accuracy gates. Use when output quality or numerics are in question — new architecture bring-up, kernel or quantization changes, KV-compression work, "model produces garbage" reports.
CLI run recipes for specific models and modes — VibeThinker, Ornith-1.0, Gemma 4 vision, image generation + upscaling, perplexity gating, whole-turn JSON-schema structured output, DSpark speculative decoding, download-model presets. Use when actually running inference for one of these.
Required workflow when adding, editing, or removing Vulkan GLSL shader consts — the committed precompiled SPIR-V table must be regenerated or tests fail on drift.
| name | implement |
| description | Drive the architect→workhorse flow — turn an approved plan into delegated Sonnet implementation with test and review gates, keeping the main (Opus/Fable) thread as coordinator. |
| disable-model-invocation | true |
| argument-hint | ["spec","plan-file path","or task description"] |
You (the main thread, running the strong model) act as architect and coordinator.
Delegate the typing to the implementer agent (Sonnet); keep design judgment and
final arbitration here. Do not make non-trivial edits yourself in this flow.
docs/*-plan.md, an issue), use it. Otherwise spawn the
architect agent to produce one, review it yourself, and fix gaps before
proceeding. A spec must name files, changes, constraints, and test filters.docs/ if it should be
committed) so it survives context compaction and can be handed to agents verbatim.implementer agent per independent work package —
in parallel only when packages touch disjoint files. Paste the full relevant
spec section into each prompt; agents see nothing else. Ordered packages run
sequentially, feeding each agent the previous report.test-runner with the spec's test filter(s). On failures,
send the digest back to the responsible implementer (SendMessage to continue
it with context intact) rather than fixing in the main thread.code-reviewer
on the diff. Triage its findings yourself — you are the arbiter; drop false
positives, route real ones back to the implementer.dotnet test suite in the main thread; that's what
test-runner is for.