Files
claude-code/plugins/accept-with-feedback/hooks/accept_with_feedback.py
Claude 411f17c228 feat: Add accept-with-feedback plugin
Add a new plugin that allows users to approve permission requests while
providing guidance to Claude. This addresses the request for an "accept
with feedback" option (similar to the existing "reject with feedback"
for plans).

The plugin includes:
- PermissionRequest hook to intercept permission prompts
- /accept-feedback command for one-time feedback
- /configure-feedback command for persistent rules
- Configuration file support for automated feedback rules

Users can now provide context like "make sure to add error handling"
when approving operations, and Claude receives this as guidance.
2025-12-14 16:13:27 +00:00

153 lines
4.5 KiB
Python

#!/usr/bin/env python3
"""
Accept with Feedback Hook for Claude Code
This hook intercepts permission requests and allows users to provide feedback
when accepting operations. The feedback is passed to Claude as a system message,
giving users a way to provide guidance while still approving the operation.
Usage:
1. Set pending feedback: /accept-feedback "your guidance here"
2. When a permission prompt appears, if pending feedback exists:
- The operation is automatically approved
- Your feedback is sent to Claude as guidance
3. Or configure always-on feedback rules in .claude/accept-feedback.json
"""
import json
import os
import sys
from datetime import datetime
from pathlib import Path
def get_config_paths():
"""Get paths for config and state files."""
claude_dir = Path.home() / ".claude"
project_dir = Path(os.environ.get("CLAUDE_PROJECT_DIR", "."))
return {
"user_config": claude_dir / "accept-feedback.json",
"project_config": project_dir / ".claude" / "accept-feedback.json",
"pending_feedback": claude_dir / "pending-accept-feedback.json",
}
def load_json_file(path):
"""Load JSON from file, return empty dict if not found."""
try:
if path.exists():
with open(path, "r") as f:
return json.load(f)
except (json.JSONDecodeError, IOError):
pass
return {}
def get_pending_feedback(session_id):
"""Get pending feedback for this session and clear it."""
paths = get_config_paths()
pending_file = paths["pending_feedback"]
pending = load_json_file(pending_file)
# Check for session-specific pending feedback
if session_id in pending:
feedback = pending[session_id]
# Clear the pending feedback after use
del pending[session_id]
try:
if pending:
with open(pending_file, "w") as f:
json.dump(pending, f)
else:
pending_file.unlink(missing_ok=True)
except IOError:
pass
return feedback.get("message"), feedback.get("one_time", True)
return None, True
def get_configured_feedback(tool_name, tool_input):
"""Get feedback configured for this tool type."""
paths = get_config_paths()
# Load both user and project configs
user_config = load_json_file(paths["user_config"])
project_config = load_json_file(paths["project_config"])
# Project config takes precedence
config = {**user_config, **project_config}
rules = config.get("rules", [])
for rule in rules:
# Check if rule matches this tool
matcher = rule.get("matcher", "*")
if matcher == "*" or tool_name in matcher.split("|"):
# Check conditions if specified
conditions = rule.get("conditions", {})
matches = True
for key, pattern in conditions.items():
value = str(tool_input.get(key, ""))
if pattern not in value:
matches = False
break
if matches:
return rule.get("feedback")
return None
def main():
"""Main hook function."""
# Read input from stdin
try:
raw_input = sys.stdin.read()
input_data = json.loads(raw_input)
except json.JSONDecodeError:
# Can't parse input, don't interfere
sys.exit(0)
session_id = input_data.get("session_id", "default")
tool_name = input_data.get("tool_name", "")
tool_input = input_data.get("tool_input", {})
# Check for pending feedback first (from /accept-feedback command)
pending_message, _ = get_pending_feedback(session_id)
if pending_message:
# We have pending feedback - approve with the feedback as system message
result = {
"hookSpecificOutput": {
"permissionDecision": "allow"
},
"systemMessage": f"[User Feedback on Approval]: {pending_message}"
}
print(json.dumps(result))
sys.exit(0)
# Check for configured feedback rules
configured_feedback = get_configured_feedback(tool_name, tool_input)
if configured_feedback:
# We have configured feedback for this tool type
result = {
"hookSpecificOutput": {
"permissionDecision": "allow"
},
"systemMessage": f"[User Guidance]: {configured_feedback}"
}
print(json.dumps(result))
sys.exit(0)
# No feedback configured - don't interfere with the normal permission flow
sys.exit(0)
if __name__ == "__main__":
main()