Build an AI Blue-Green Deployment Risk Scorer with Claude API

pip install anthropic PyGithub requests

import anthropic
from github import Github
 
client = anthropic.Anthropic()
 
 
def gather_risk_signals(repo_name: str, pr_number: int, github_token: str) -> dict:
    gh = Github(github_token)
    repo = gh.get_repo(repo_name)
    pr = repo.get_pull(pr_number)
 
    changed_files = list(pr.get_files())
 
    return {
        "files_changed": len(changed_files),
        "lines_changed": sum(f.additions + f.deletions for f in changed_files),
        "touches_database_migration": any("migrations/" in f.filename for f in changed_files),
        "touches_critical_path": any(f.filename in CRITICAL_FILES for f in changed_files),
        "test_coverage_delta": get_coverage_delta(pr),
        "ci_status": get_ci_check_results(repo, pr),
        "similar_past_deploys": find_similar_past_deploys(changed_files),
    }
 
 
CRITICAL_FILES = ["src/payments/", "src/auth/", "src/checkout/"]
 
 
def find_similar_past_deploys(changed_files: list) -> list[dict]:
    """Look at deploy history for past changes to these same files —
    did previous deploys touching this code cause incidents?"""
    touched_paths = [f.filename for f in changed_files]
    return query_deploy_history_for_paths(touched_paths, months=6)

import json
 
SCORE_PROMPT = """Score the cutover risk for this blue-green deployment.
 
PR summary: {pr_title}
Files changed: {files_changed}, Lines changed: {lines_changed}
Touches database migration: {touches_migration}
Touches critical path (payments/auth/checkout): {touches_critical}
Test coverage delta: {coverage_delta}
CI status: {ci_status}
 
Historical context — past deploys touching similar files:
{similar_deploys}
 
Unlike a canary, blue-green cutover shifts 100% of traffic at once — there
is no gradual traffic-based warning before full exposure. Score risk
accordingly, weighing:
1. Blast radius if this deploy has a bug (critical path = high blast radius)
2. Whether a database migration is involved (harder/riskier to instant-rollback
   if the new schema is already being written to)
3. Whether test coverage genuinely covers the changed code, not just exists somewhere in the repo
4. Whether similar past changes to this code caused incidents
 
Respond with ONLY valid JSON:
{{"risk_score": "low"|"medium"|"high", "reasoning": "...",
  "recommended_action": "proceed_with_cutover" | "recommend_canary_instead" | "recommend_additional_testing_first"}}"""
 
 
def score_deployment_risk(signals: dict, pr_title: str) -> dict:
    response = client.messages.create(
        model="claude-sonnet-5",
        max_tokens=800,
        messages=[{
            "role": "user",
            "content": SCORE_PROMPT.format(
                pr_title=pr_title, files_changed=signals["files_changed"],
                lines_changed=signals["lines_changed"],
                touches_migration=signals["touches_database_migration"],
                touches_critical=signals["touches_critical_path"],
                coverage_delta=signals["test_coverage_delta"],
                ci_status=signals["ci_status"],
                similar_deploys=signals["similar_past_deploys"],
            )
        }]
    )
    text = response.content[0].text.strip()
    if text.startswith("```"):
        text = text.split("```")[1].replace("json", "", 1).strip()
    return json.loads(text)