菜单
Use when executing implementation plans with independent tasks in the current session
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
Use when setting up the lrt-rocm workspace for the first time, or when a new team member installs the plugin
Use when starting any conversation - establishes how to route tasks to the right skills using the routing table
Use when building CLR (HIP/OCL) with PAL and/or ROCr backends on Windows using rocm-systems — covers prerequisites, CMake flags, build configurations, and troubleshooting for d3dumddi.h and Clang discovery issues
Use when implementation is complete, all tests pass, and you need to decide how to integrate the work - guides completion of development work by presenting structured options for merge, PR, or cleanup
Use when starting feature work that needs isolation from current workspace or before executing implementation plans - creates isolated git worktrees with smart directory selection and safety verification
Use when you have a spec or requirements for a multi-step task, before touching code
| name | subagent-driven-development |
| description | Use when executing implementation plans with independent tasks in the current session |
Execute plan by dispatching registered agents per task, with two-stage review after each: spec compliance first, then code quality.
Core principle: Fresh agent per task + two-stage review (spec then quality) = high quality, fast iteration
| Agent | Purpose |
|---|---|
lrt-rocm:implementer | Implements a single task — writes code, tests, commits, self-reviews |
lrt-rocm:spec-reviewer | Verifies implementation matches spec (read-only, doesn't trust implementer's report) |
lrt-rocm:code-reviewer | Reviews code quality, architecture, production readiness |
digraph process {
rankdir=TB;
subgraph cluster_per_task {
label="Per Task";
"Dispatch lrt-rocm:implementer" [shape=box];
"Implementer asks questions?" [shape=diamond];
"Answer questions, provide context" [shape=box];
"Implementer implements, tests, commits, self-reviews" [shape=box];
"Dispatch lrt-rocm:spec-reviewer" [shape=box];
"Spec reviewer confirms code matches spec?" [shape=diamond];
"Implementer fixes spec gaps" [shape=box];
"Dispatch lrt-rocm:code-reviewer" [shape=box];
"Code reviewer approves?" [shape=diamond];
"Implementer fixes quality issues" [shape=box];
"Mark task complete" [shape=box];
}
"Read plan, extract all tasks with full text, create task list" [shape=box];
"More tasks remain?" [shape=diamond];
"Dispatch final lrt-rocm:code-reviewer for entire implementation" [shape=box];
"Use lrt-rocm:finishing-a-development-branch" [shape=box style=filled fillcolor=lightgreen];
"Read plan, extract all tasks with full text, create task list" -> "Dispatch lrt-rocm:implementer";
"Dispatch lrt-rocm:implementer" -> "Implementer asks questions?";
"Implementer asks questions?" -> "Answer questions, provide context" [label="yes"];
"Answer questions, provide context" -> "Dispatch lrt-rocm:implementer";
"Implementer asks questions?" -> "Implementer implements, tests, commits, self-reviews" [label="no"];
"Implementer implements, tests, commits, self-reviews" -> "Dispatch lrt-rocm:spec-reviewer";
"Dispatch lrt-rocm:spec-reviewer" -> "Spec reviewer confirms code matches spec?";
"Spec reviewer confirms code matches spec?" -> "Implementer fixes spec gaps" [label="no"];
"Implementer fixes spec gaps" -> "Dispatch lrt-rocm:spec-reviewer" [label="re-review"];
"Spec reviewer confirms code matches spec?" -> "Dispatch lrt-rocm:code-reviewer" [label="yes"];
"Dispatch lrt-rocm:code-reviewer" -> "Code reviewer approves?";
"Code reviewer approves?" -> "Implementer fixes quality issues" [label="no"];
"Implementer fixes quality issues" -> "Dispatch lrt-rocm:code-reviewer" [label="re-review"];
"Code reviewer approves?" -> "Mark task complete" [label="yes"];
"Mark task complete" -> "More tasks remain?";
"More tasks remain?" -> "Dispatch lrt-rocm:implementer" [label="yes"];
"More tasks remain?" -> "Dispatch final lrt-rocm:code-reviewer for entire implementation" [label="no"];
"Dispatch final lrt-rocm:code-reviewer for entire implementation" -> "Use lrt-rocm:finishing-a-development-branch";
}
Agent tool:
subagent_type: "lrt-rocm:implementer"
description: "Implement Task N: [task name]"
prompt: |
## Task Description
[FULL TEXT of task from plan — paste it, don't make agent read file]
## Context
[Where this fits, dependencies, architectural context]
Work from: [directory]
Agent tool:
subagent_type: "lrt-rocm:spec-reviewer"
description: "Review spec compliance for Task N"
prompt: |
## What Was Requested
[FULL TEXT of task requirements]
## What Implementer Claims They Built
[From implementer's report]
Agent tool:
subagent_type: "lrt-rocm:code-reviewer"
description: "Code quality review for Task N"
prompt: |
WHAT_WAS_IMPLEMENTED: [from implementer's report]
PLAN_OR_REQUIREMENTS: Task N from [plan-file]
BASE_SHA: [commit before task]
HEAD_SHA: [current commit]
DESCRIPTION: [task summary]
Additional checks:
- Does each file have one clear responsibility?
- Is the implementation following the file structure from the plan?
- Did this change create overly large files?
Agents default to opus. Override with the model parameter for cost savings:
model: "haiku"model: "sonnet"DONE: Proceed to spec compliance review.
DONE_WITH_CONCERNS: Read the concerns. If about correctness or scope, address before review. If observations (e.g., "this file is getting large"), note and proceed.
NEEDS_CONTEXT: Provide the missing context and re-dispatch.
BLOCKED: Assess the blocker:
Never ignore an escalation or force retry without changes.
Never:
If reviewer finds issues:
If agent fails task:
Required workflow skills: