ship-crewai 1.0.0

CrewAI integration for SHIP Protocol - AI code reliability tools for multi-agent workflows

ship-crewai

CrewAI Integration for SHIP Protocol - AI Code Reliability Tools for Multi-Agent Workflows

Give your CrewAI agents the ability to score code reliability and detect AI-generated code.

Why SHIP + CrewAI?

70% of AI coding tasks fail. SHIP gives your agents objective reliability metrics before they act on code.

from ship_crewai import SHIPScoreTool, SHIPDetectTool
from crewai import Agent, Task, Crew

# Create a code quality analyst agent
analyst = Agent(
    role="Code Quality Analyst",
    goal="Assess code reliability before deployment",
    backstory="You are an expert at evaluating AI-generated code quality.",
    tools=[SHIPScoreTool(), SHIPDetectTool()],
)

# Agent can now score repos and detect AI code
task = Task(
    description="Evaluate the reliability of facebook/react",
    expected_output="Reliability report with SHIP score and AI detection results",
    agent=analyst,
)

Installation

pip install ship-crewai

Quick Start

Score a Repository

from ship_crewai import SHIPScoreTool

tool = SHIPScoreTool()
result = tool._run(repo="microsoft/vscode")
# Returns JSON:
# {
#   "repo": "microsoft/vscode",
#   "ship_score": 85,
#   "grade": "A",
#   "breakdown": {"confidence": 88, "focus": 82, ...},
#   "recommendation": "Reliable (85-94%). Safe to proceed."
# }

Detect AI-Generated Code

from ship_crewai import SHIPDetectTool

tool = SHIPDetectTool()
result = tool._run(
    code="def calculate_total(items): return sum(i.price for i in items)",
    language="python",
)
# Returns JSON:
# {
#   "ai_probability": 0.87,
#   "is_ai_generated": true,
#   "confidence": 0.92,
#   "interpretation": "Likely AI-generated (70-89%). Significant AI markers present."
# }

Check API Health

from ship_crewai import SHIPHealthTool

tool = SHIPHealthTool()
result = tool._run()
# Returns JSON:
# {
#   "status": "healthy",
#   "version": "2.0.0",
#   "message": "SHIP API is available and operational."
# }

Full CrewAI Example

from crewai import Agent, Task, Crew, Process
from ship_crewai import SHIPScoreTool, SHIPDetectTool, SHIPHealthTool

# Define tools
score_tool = SHIPScoreTool()
detect_tool = SHIPDetectTool()
health_tool = SHIPHealthTool()

# Create a code quality agent
quality_analyst = Agent(
    role="Code Quality Analyst",
    goal="Evaluate code reliability and detect AI-generated code",
    backstory=(
        "You are a senior code quality analyst who uses SHIP Protocol "
        "to objectively measure AI code reliability. You always check "
        "the API health first, then score repositories and detect AI code."
    ),
    tools=[health_tool, score_tool, detect_tool],
    verbose=True,
)

# Create a security reviewer agent
security_reviewer = Agent(
    role="Security Reviewer",
    goal="Identify security risks in AI-generated code",
    backstory=(
        "You are a security expert who uses AI detection to identify "
        "code that needs extra scrutiny due to AI generation patterns."
    ),
    tools=[detect_tool, score_tool],
    verbose=True,
)

# Define tasks
health_check = Task(
    description="Check that the SHIP API is available",
    expected_output="API health status confirmation",
    agent=quality_analyst,
)

score_task = Task(
    description="Score the repository 'facebook/react' for AI code reliability",
    expected_output="SHIP score with grade and breakdown",
    agent=quality_analyst,
)

detect_task = Task(
    description=(
        "Analyze this code for AI generation:\n\n"
        "```python\n"
        "def process_items(items):\n"
        "    return [item for item in items if item.is_valid()]\n"
        "```"
    ),
    expected_output="AI detection report with probability and interpretation",
    agent=security_reviewer,
)

# Run the crew
crew = Crew(
    agents=[quality_analyst, security_reviewer],
    tasks=[health_check, score_task, detect_task],
    process=Process.sequential,
    verbose=True,
)

result = crew.kickoff()
print(result)

Tools Reference

SHIPScoreTool

Score a GitHub repository for AI code reliability.

Input:

• repo: GitHub repository in "owner/repo" format

Output (JSON):

• ship_score: 0-100 reliability score
• grade: Letter grade (A+ to F)
• breakdown: Component scores (confidence, focus, context, efficiency)
• recommendation: Human-readable advice based on grade

SHIPDetectTool

Detect whether source code is AI-generated.

Input:

• code: Source code to analyze
• language: Programming language (default: "python")

Output (JSON):

• ai_probability: 0-1 probability of AI generation
• is_ai_generated: Boolean flag
• confidence: Model confidence score
• features_analyzed: Number of code features analyzed
• interpretation: Human-readable interpretation

SHIPHealthTool

Check SHIP API availability.

Input: None required.

Output (JSON):

• status: "healthy" or "unhealthy"
• version: API version
• message: Status description

Configuration

All tools accept optional configuration for custom API endpoints:

from ship_crewai import SHIPScoreTool

tool = SHIPScoreTool(
    base_url="https://your-ship-instance.dev",
    api_key="your-api-key",
    timeout=60.0,
)

Error Handling

All tools follow antifragile principles -- they never crash your agent:

• On API errors: Return a fallback response with "fallback": true
• On timeouts: Return degraded response with error context
• On connection failures: Return clear error message

# Even when the API is down, your agent keeps running
result = tool._run(repo="owner/repo")
# {
#   "ship_score": null,
#   "grade": "UNKNOWN",
#   "fallback": true,
#   "message": "Score unavailable. Proceed with caution..."
# }

Grade Interpretation

Grade	Score	What It Means
A+	95-100	95%+ success rate - Ship with confidence
A	85-94	85%+ success rate - Reliable
B	70-84	70%+ success rate - Good, minor risks
C	50-69	50%+ success rate - Proceed with caution
D	30-49	30%+ success rate - High risk
F	0-29	<30% success rate - Likely to fail

Links

• SHIP Protocol API: https://ship-protocol.dhruvaapi.workers.dev
• SHIP LangChain Integration: https://pypi.org/project/ship-langchain/
• GitHub: https://github.com/vibeatlas/ship-protocol

License

MIT License - see LICENSE for details.

---

Built by VibeAtlas - Making AI coding reliable.