π The fast, Pythonic way to build MCP servers and clients.
#!/usr/bin/env python
# /// script
# requires-python = ">=3.10"
# dependencies = [
# "httpx",
# ]
# ///
"""
Auto-close duplicate GitHub issues.
This script runs on a schedule to automatically close issues that have been
marked as duplicates and haven't received any preventing activity.
"""
import json
import os
from dataclasses import dataclass
from datetime import UTC, datetime, timedelta
from pathlib import Path
import httpx
MARVIN_BOT_ID = 225465937
CLOSURE_NOTICE = (
"Closing this issue as a duplicate based on the automated analysis above."
)
ESTABLISHED_ASSOCIATIONS = {"CONTRIBUTOR", "COLLABORATOR", "MEMBER", "OWNER"}
@dataclass
class Issue:
"""Represents a GitHub issue."""
number: int
title: str
state: str
created_at: str
user_id: int
user_login: str
@dataclass
class Comment:
"""Represents a GitHub comment."""
id: int
body: str
created_at: str
user_id: int
user_login: str
user_type: str
@dataclass
class Reaction:
"""Represents a reaction on a comment."""
user_id: int
user_login: str
content: str
class GitHubClient:
"""Client for interacting with GitHub API."""
def __init__(self, token: str, owner: str, repo: str):
self.token = token
self.owner = owner
self.repo = repo
self.headers = {
"Authorization": f"token {token}",
"Accept": "application/vnd.github.v3+json",
}
self.base_url = f"https://api.github.com/repos/{owner}/{repo}"
def get_potential_duplicate_issues(self) -> list[Issue]:
"""Fetch open issues with the potential-duplicate label."""
url = f"{self.base_url}/issues"
issues = []
with httpx.Client() as client:
page = 1
while page <= 10: # Safety limit
response = client.get(
url,
headers=self.headers,
params={
"state": "open",
"labels": "potential-duplicate",
"per_page": 100,
"page": page,
},
)
if response.status_code != 200:
print(f"Error fetching issues: {response.status_code}")
break
data = response.json()
if not data:
break
for item in data:
# Skip pull requests
if "pull_request" in item:
continue
issues.append(
Issue(
number=item["number"],
title=item["title"],
state=item["state"],
created_at=item["created_at"],
user_id=item["user"]["id"],
user_login=item["user"]["login"],
)
)
page += 1
return issues
def _read_all(self, path: str) -> list[dict]:
records = []
with httpx.Client() as client:
for page in range(1, 11):
response = client.get(
f"{self.base_url}/{path}",
headers=self.headers,
params={"page": page, "per_page": 100},
)
response.raise_for_status()
batch = response.json()
if not isinstance(batch, list):
raise ValueError("Expected a GitHub collection")
records.extend(batch)
if "next" not in response.links:
return records
raise ValueError("GitHub collection exceeded the ten-page limit")
def get_issue_comments(self, issue_number: int) -> list[Comment]:
return [
Comment(
id=comment["id"],
body=comment["body"],
created_at=comment["created_at"],
user_id=comment["user"]["id"],
user_login=comment["user"]["login"],
user_type=comment["user"]["type"],
)
for comment in self._read_all(f"issues/{issue_number}/comments")
]
def get_comment_reactions(self, comment_id: int) -> list[Reaction]:
return [
Reaction(
user_id=reaction["user"]["id"],
user_login=reaction["user"]["login"],
content=reaction["content"],
)
for reaction in self._read_all(f"issues/comments/{comment_id}/reactions")
]
def get_issue_timeline(self, issue_number: int) -> list[dict]:
return self._read_all(f"issues/{issue_number}/timeline")
def reopen_duplicate(self, issue_number: int, closed_at: str) -> bool:
url = f"{self.base_url}/issues/{issue_number}"
with httpx.Client() as client:
response = client.get(url, headers=self.headers)
response.raise_for_status()
issue = response.json()
if not is_closed_duplicate(issue) or issue["closed_at"] != closed_at:
return False
response = client.patch(url, headers=self.headers, json={"state": "open"})
response.raise_for_status()
return response.json()["state"] == "open"
def remove_label(self, issue_number: int, label: str) -> bool:
"""Remove a label from an issue."""
url = f"{self.base_url}/issues/{issue_number}/labels/{label}"
with httpx.Client() as client:
response = client.delete(url, headers=self.headers)
return response.status_code in [200, 204]
def close_issue(self, issue_number: int, comment: str) -> bool:
"""Close an issue with a comment and add duplicate label."""
# First add the comment
comment_url = f"{self.base_url}/issues/{issue_number}/comments"
with httpx.Client() as client:
response = client.post(
comment_url, headers=self.headers, json={"body": comment}
)
if response.status_code != 201:
print(f"Failed to add comment to issue #{issue_number}")
return False
# Swap labels: remove potential-duplicate, add duplicate
self.remove_label(issue_number, "potential-duplicate")
labels_url = f"{self.base_url}/issues/{issue_number}/labels"
with httpx.Client() as client:
response = client.post(
labels_url, headers=self.headers, json={"labels": ["duplicate"]}
)
if response.status_code not in [200, 201]:
print(f"Failed to add duplicate label to issue #{issue_number}")
# Then close the issue
issue_url = f"{self.base_url}/issues/{issue_number}"
with httpx.Client() as client:
response = client.patch(
issue_url, headers=self.headers, json={"state": "closed"}
)
return response.status_code == 200
def find_duplicate_comment(comments: list[Comment]) -> Comment | None:
"""Find a bot comment marking the issue as duplicate."""
for comment in comments:
# Check for the specific duplicate message format from a bot. With
# one duplicate the comment says "issue", not "issues".
if (
comment.user_type == "Bot"
and "possible duplicate issue" in comment.body.lower()
):
return comment
return None
def was_already_auto_closed(comments: list[Comment]) -> bool:
"""Check if this issue was already auto-closed once."""
for comment in comments:
if (
comment.user_type == "Bot"
and "closing this issue as a duplicate" in comment.body.lower()
):
return True
return False
def is_past_cooldown(duplicate_comment: Comment) -> bool:
"""Check if the 3-day cooldown period has passed."""
comment_date = datetime.fromisoformat(
duplicate_comment.created_at.replace("Z", "+00:00")
)
three_days_ago = datetime.now(UTC) - timedelta(days=3)
return comment_date <= three_days_ago
def has_human_activity(
issue: Issue,
duplicate_comment: Comment,
all_comments: list[Comment],
reactions: list[Reaction],
) -> bool:
"""Check if there's human activity that should prevent auto-closure."""
comment_date = datetime.fromisoformat(
duplicate_comment.created_at.replace("Z", "+00:00")
)
# Check for preventing reactions (thumbs down)
for reaction in reactions:
if reaction.content in ["-1", "confused"]:
print(
f"Issue #{issue.number}: Has preventing reaction from {reaction.user_login}"
)
return True
# Check for any human comment after the duplicate marking
for comment in all_comments:
comment_date_check = datetime.fromisoformat(
comment.created_at.replace("Z", "+00:00")
)
if comment_date_check > comment_date:
if comment.user_type != "Bot":
print(
f"Issue #{issue.number}: {comment.user_login} commented after duplicate marking"
)
return True
return False
def is_closed_duplicate(issue: dict) -> bool:
return (
issue["state"] == "closed"
and "pull_request" not in issue
and any(label["name"] == "duplicate" for label in issue["labels"])
)
def reconsider_duplicate(client: GitHubClient, event: dict) -> None:
issue = event["issue"]
comment = event["comment"]
author = comment["user"]
if not is_closed_duplicate(issue) or author["type"] != "User":
return
if (
author["id"] != issue["user"]["id"]
and comment["author_association"] not in ESTABLISHED_ASSOCIATIONS
):
return
closed_at = issue["closed_at"]
if not closed_at or comment["created_at"] < closed_at:
return
number = issue["number"]
closed_time = datetime.fromisoformat(closed_at)
found_close = False
for entry in reversed(client.get_issue_timeline(number)):
kind, created_at = entry["event"], entry.get("created_at")
if kind in {"closed", "reopened"}:
if (
found_close
or kind != "closed"
or created_at != closed_at
or (entry.get("actor") or {}).get("id") != MARVIN_BOT_ID
):
return
found_close = True
elif (
found_close
and kind == "commented"
and (entry.get("user") or {}).get("id") == MARVIN_BOT_ID
and (entry.get("body") or "").startswith(CLOSURE_NOTICE)
and created_at
and timedelta(0)
<= closed_time - datetime.fromisoformat(created_at)
<= timedelta(minutes=5)
):
if client.reopen_duplicate(number, closed_at):
print(
f"Reopened #{number} for maintainer review after comment {comment['id']}"
)
return
def main():
"""Main entry point for auto-closing duplicate issues."""
print("[DEBUG] Starting auto-close duplicates script")
# Get environment variables
token = os.environ.get("GITHUB_TOKEN")
if not token:
raise ValueError("GITHUB_TOKEN environment variable is required")
owner = os.environ.get("GITHUB_REPOSITORY_OWNER", "prefecthq")
repo = os.environ.get("GITHUB_REPOSITORY_NAME", "fastmcp")
print(f"[DEBUG] Repository: {owner}/{repo}")
# Initialize client
client = GitHubClient(token, owner, repo)
if os.environ.get("GITHUB_EVENT_NAME") == "issue_comment":
event = json.loads(Path(os.environ["GITHUB_EVENT_PATH"]).read_text())
reconsider_duplicate(client, event)
return
# Only fetch issues with the potential-duplicate label
all_issues = client.get_potential_duplicate_issues()
print(f"[DEBUG] Found {len(all_issues)} open issues with potential-duplicate label")
closed_count = 0
cleared_count = 0
failed_reads = 0
for issue in all_issues:
try:
comments = client.get_issue_comments(issue.number)
except (httpx.HTTPError, ValueError, KeyError, TypeError) as error:
print(f"Unable to read comments for #{issue.number}: {error}")
failed_reads += 1
continue
# Look for duplicate marking comment
duplicate_comment = find_duplicate_comment(comments)
if not duplicate_comment:
# Label exists but no bot comment - clean up the label
print(f"[DEBUG] Issue #{issue.number} has label but no duplicate comment")
client.remove_label(issue.number, "potential-duplicate")
cleared_count += 1
continue
# Skip if already auto-closed once (someone reopened it intentionally)
if was_already_auto_closed(comments):
print(
f"[DEBUG] Issue #{issue.number} was already auto-closed, removing label"
)
client.remove_label(issue.number, "potential-duplicate")
cleared_count += 1
continue
print(f"[DEBUG] Issue #{issue.number} has duplicate comment")
# Still in cooldown period - skip for now, check again later
if not is_past_cooldown(duplicate_comment):
print(f"[DEBUG] Issue #{issue.number} still in 3-day cooldown period")
continue
try:
reactions = client.get_comment_reactions(duplicate_comment.id)
except (httpx.HTTPError, ValueError, KeyError, TypeError) as error:
print(f"Unable to read reactions for #{issue.number}: {error}")
failed_reads += 1
continue
# Check for human activity that prevents closure
if has_human_activity(issue, duplicate_comment, comments, reactions):
print(f"[DEBUG] Issue #{issue.number} has human activity, removing label")
client.remove_label(issue.number, "potential-duplicate")
cleared_count += 1
continue
# No human activity after cooldown - close as duplicate
close_message = (
f"{CLOSURE_NOTICE}\n\n"
"The duplicate issues identified contain existing discussions and potential solutions. "
"Please add your π to those issues if they match your use case.\n\n"
"If this was closed in error, please leave a comment explaining why this is not "
"a duplicate and we'll reopen it."
)
if client.close_issue(issue.number, close_message):
print(f"[SUCCESS] Closed issue #{issue.number} as duplicate")
closed_count += 1
else:
print(f"[ERROR] Failed to close issue #{issue.number}")
print(
f"[DEBUG] Processing complete. Closed {closed_count} duplicates, "
f"cleared {cleared_count} from review, {failed_reads} evidence reads failed"
)
if failed_reads:
raise SystemExit(1)
if __name__ == "__main__":
main()