菜单
An agent-driven PR review workflow that constrains step-by-step execution: the agent triages PR scope, branches by review depth, invokes code-reviewer skills, and terminates with approve or request-changes — all guided by a declarative state machine.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/loonghao/agentverse --skill workflow-ci-review-pipeline命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
An autonomous code-assistant agent that can review code, answer technical questions, generate tests, and suggest refactors — powered by AgentVerse skills and an empathetic developer soul. Exposes MCP tools and is A2A (Agent-to-Agent) compatible.
Publish, discover, and manage AI skills, agents, workflows, souls and prompts from the AgentVerse marketplace. Use when working with the agentverse CLI to search/publish artifacts, authenticate, or manage AI agent ecosystem components.
Runs a suite of smoke tests against any REST API and reports status, latency, and failure details.
基于 SOC 职业分类
正在显示 SKILL.md
| name | workflow-ci-review-pipeline |
| kind | workflow |
| description | An agent-driven PR review workflow that constrains step-by-step execution: the agent triages PR scope, branches by review depth, invokes code-reviewer skills, and terminates with approve or request-changes — all guided by a declarative state machine. |
| tags | ["workflow","agent","orchestration","state-machine","branching","code-review"] |
| version | 0.1.0 |
| author | agentverse |
| license | MIT |
| metadata | {"openclaw":{"homepage":"https://github.com/loonghao/agentverse","emoji":"🔄"}} |
A workflow artifact defines the step-by-step orchestration logic that constrains how an Agent executes. It is a declarative state machine: each step tells the agent what to do, what to write into the shared context, and which condition determines the next step. The agent is the executor — the workflow is the script it must follow.
| Concept | Description |
|---|---|
entry | The first step the agent executes |
context | Typed shared state — agents read from and write to it across all steps |
decision | A step where the agent reasons and sets context keys (no external skill) |
skill | A step that invokes an AgentVerse skill and writes outputs to context |
transitions | Conditional routing rules — evaluated in order after each step |
writes | Declares which context keys a step is allowed to modify |
__end__ | Special goto target that terminates the workflow |
workflow.toml)[package]
kind = "workflow"
namespace = "agentverse"
name = "ci-review-pipeline"
description = "Agent-driven PR review: triage → branch by depth → approve or request-changes"
# ── Entry & global settings ───────────────────────────────────────────────────
[workflow]
entry = "triage"
timeout = "30m"
# ── Shared context (typed state the agent reads/writes) ───────────────────────
[workflow.context]
pr_url = { type = "string", required = }
= { type = , default = , enum = [, ] }
= { type = , default = }
= { type = , default = [] }
= { type = , default = }
=
=
=
=
= []
=
=
=
=
=
=
=
=
= { diff = , rules = [, , , ] }
= [, ]
=
= { max_attempts = , back = }
=
=
=
=
=
=
=
=
= { diff = , rules = [] }
= [, ]
=
=
=
=
=
= { message = , approve = }
= []
=
=
=
=
=
= { issues = , score = , approve = }
=
= [, , , , ]
=
=
┌──────────┐ depth='deep' ┌─────────────┐ score>=80 ┌─────────┐
│ triage │────────────────▶│ full-review │────────────▶│ approve │──▶ __end__
│(decision)│ │ (skill) │ score<80 └─────────┘
│ │ └─────────────┘────────────▶┌──────────────────┐
│ │ depth='shallow' ┌─────────────┐ │ request-changes │──▶ __end__
│ │────────────────▶│ quick-check │────────────▶└──────────────────┘
└──────────┘ │ (skill) │ (unconditional)
└─────────────┘
The agent follows this state machine step by step:
triage — agent reasons about PR size and writes depth to contextfull-review or quick-check based on depthscore and issues to contextapprove or request-changes based on score__end__content.json){
"schema_version": "1.0",
"kind": "workflow",
"entry": "triage",
"context": {
"pr_url": { "type": "string", "required": true },
"depth": { "type": "string", "default": "shallow", "enum": ["shallow","deep"] },
"score": { "type": "integer", "default": 0 },
"issues": { "type": "array", "default": [] },
"approved": { "type": "boolean", "default": false }
},
"steps": [
{
"id": "triage", "kind": "decision",
"instruction": "Count changed lines in {{context.pr_url}}. Set depth='deep' if >500.",
"writes": ["depth"],
"transitions": [
{ "when": "context.depth == 'deep'", "goto": "full-review" },
{ "when": "context.depth == 'shallow'", "goto": "quick-check" }
]
},
{
"id": "full-review", "kind": "skill",
"use": "agentverse-ci/code-reviewer@>=0.1.0",
"inputs": { "diff": "{{context.pr_url}}", "rules": ["security","correctness","performance","style"] },
"writes": ["score","issues"],
"on_error": "retry", "retry": { "max_attempts": 2, "backoff": "30s" },
"transitions": [
{ "when": "context.score >= 80", "goto": "approve" },
{ "when": "context.score < 80", "goto": "request-changes" }
]
},
{
"id": "quick-check", "kind": "skill",
"use": "agentverse-ci/code-reviewer@>=0.1.0",
"inputs": { "diff": "{{context.pr_url}}", "rules": ["style"] },
"writes": ["score","issues"],
"transitions": [{ "goto": "approve" }]
},
{
"id": "approve", "kind": "skill",
"use": "agentverse-ci/pr-commenter@>=0.1.0",
"inputs": { "message": "✅ Review passed (score: {{context.score}})", "approve": true },
"writes": ["approved"],
"transitions": [{ "goto": "__end__" }]
},
{
"id": "request-changes", "kind": "skill",
"use": "agentverse-ci/pr-commenter@>=0.1.0",
"inputs": { "issues": "{{context.issues}}", "score": "{{context.score}}", "approve": false },
"transitions": [{ "goto": "__end__" }]
}
]
}
# Publish the workflow
agentverse publish --file workflow.toml
# → Published workflow agentverse/ci-review-pipeline@0.1.0
# Run with initial context (agent executes step by step)
agentverse run --kind workflow --namespace agentverse --name ci-review-pipeline \
--context pr_url=https://github.com/org/repo/pull/42
| Framework | Mapping |
|---|---|
| LangGraph | Steps → nodes; transitions → conditional edges; context → state |
| CrewAI Flows | Steps → @listen/@router; context → self.state |
| OpenAI Swarm | Steps → routines; agent steps → handoffs |
| Prefect | Steps → tasks; parallel → asyncio.gather; loop → retry |
writes is enforced — a step cannot touch context keys it hasn't declared.decision steps require no external tool — the agent itself reasons and writes.when wins; omit when for a catch-all.soul artifact to give the agent a consistent persona across all steps.