workflow-ci-review-pipeline

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":"🔄"}}

Workflow: CI Review Pipeline

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.

Core Concepts

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 Manifest (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 = true }
depth    = { type = "string",  default = "shallow", enum = ["shallow", "deep"] }
score    = { type = "integer", default = 0 }
issues   = { type = "array",   default = [] }
approved = { type = "boolean", default = false }

# ── Step 1: Triage — agent DECIDES review depth ───────────────────────────────
[[workflow.steps]]
id          = "triage"
name        = "Triage PR scope"
kind        = "decision"
instruction = """
Fetch the PR at {{context.pr_url}} and count the total changed lines.
If > 500 lines or security-sensitive paths are touched, set depth = "deep".
Otherwise set depth = "shallow".
"""
writes = ["depth"]

[[workflow.steps.transitions]]
when = "context.depth == 'deep'"
goto = "full-review"

[[workflow.steps.transitions]]
when = "context.depth == 'shallow'"
goto = "quick-check"

# ── Step 2a: Full review (deep path) ─────────────────────────────────────────
[[workflow.steps]]
id       = "full-review"
name     = "Full code 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" }

[[workflow.steps.transitions]]
when = "context.score >= 80"
goto = "approve"

[[workflow.steps.transitions]]
when = "context.score < 80"
goto = "request-changes"

# ── Step 2b: Quick check (shallow path) ──────────────────────────────────────
[[workflow.steps]]
id     = "quick-check"
name   = "Style-only check"
kind   = "skill"
use    = "agentverse-ci/code-reviewer@>=0.1.0"
inputs = { diff = "{{context.pr_url}}", rules = ["style"] }
writes = ["score", "issues"]

[[workflow.steps.transitions]]
goto = "approve"

# ── Step 3a: Approve ─────────────────────────────────────────────────────────
[[workflow.steps]]
id     = "approve"
name   = "Post approval comment"
kind   = "skill"
use    = "agentverse-ci/pr-commenter@>=0.1.0"
inputs = { message = "✅ Review passed (score: {{context.score}})", approve = true }
writes = ["approved"]

[[workflow.steps.transitions]]
goto = "__end__"

# ── Step 3b: Request changes ──────────────────────────────────────────────────
[[workflow.steps]]
id     = "request-changes"
name   = "Post change requests"
kind   = "skill"
use    = "agentverse-ci/pr-commenter@>=0.1.0"
inputs = { issues = "{{context.issues}}", score = "{{context.score}}", approve = false }

[[workflow.steps.transitions]]
goto = "__end__"

[metadata]
tags     = ["workflow", "agent", "state-machine", "branching", "code-review"]
homepage = "https://github.com/loonghao/agentverse"
license  = "MIT"

State Machine Diagram

┌──────────┐  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:

1. triage — agent reasons about PR size and writes depth to context
2. Transitions route to full-review or quick-check based on depth
3. Each review step writes score and issues to context
4. Transitions route to approve or request-changes based on score
5. Terminal steps go to __end__

AgentVerse Content (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 and Run

# 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 Compatibility

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

Notes

• Context is the single source of truth — all steps read from and write to it.
• 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.
• Transitions are ordered — the first matching when wins; omit when for a catch-all.
• Combine with a soul artifact to give the agent a consistent persona across all steps.