appautomaton/latex-arxiv-SKILL / .codex/skills/collaborating-with-claude
collaborating-with-claude
VerifiedUse the Claude Code CLI to consult Claude and delegate coding tasks for prototyping, debugging, and code review. Supports multi-turn sessions via SESSION_ID. Optimized for low-token, file/line-based handoff.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.0 KB |
| assets/prompt-template.md | markdown | 1.1 KB |
| references/shell-quoting.md | markdown | 1.2 KB |
| scripts/claude_bridge.py | script | 11.8 KB |
scripts/claude_bridge.py
View on GitHub#!/usr/bin/env python3
"""
Claude Code Bridge Script for Codex Skills.
Wraps the Claude Code CLI to provide a JSON-based interface and multi-turn sessions via SESSION_ID.
"""
import argparse
import json
import shutil
import subprocess
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple
def read_output_lines(cmd: List[str], cwd: Optional[str] = None) -> Tuple[List[str], List[str], int]:
"""Execute a command and capture stdout/stderr as lists of lines."""
popen_cmd = cmd.copy()
claude_path = shutil.which("claude") or cmd[0]
popen_cmd[0] = claude_path
process = subprocess.Popen(
popen_cmd,
shell=False,
stdin=subprocess.DEVNULL,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
universal_newlines=True,
encoding="utf-8",
cwd=cwd,
)
stdout, stderr = process.communicate()
stdout_lines = stdout.splitlines() if stdout else []
stderr_lines = stderr.splitlines() if stderr else []
return stdout_lines, stderr_lines, process.returncode
def extract_text(value: Any) -> str:
"""Recursively extract text from a Claude Code JSON payload."""
if isinstance(value, str):
return value
if isinstance(value, list):
return "".join(extract_text(item) for item in value)
if isinstance(value, dict):
if isinstance(value.get("text"), str):
return value["text"]
if "content" in value:
return extract_text(value["content"])
if "delta" in value:
return extract_text(value["delta"])
if "message" in value:
return extract_text(value["message"])
return ""
def update_role(obj: Dict[str, Any], current_role: Optional[str]) -> Optional[str]:
"""Track the current message role for streaming output."""
role = obj.get("role")
if isinstance(role, str):
return role
message = obj.get("message")
if isinstance(message, dict):
role = message.get("role")
if isinstance(role, str):
return role
return current_role
def parse_stream_json(lines: List[str]) -> Tuple[Optional[str], str, List[Dict[str, Any]], str]:
"""Parse stream-json output into session_id, agent_messages, all_messages, error."""
all_messages: List[Dict[str, Any]] = []
agent_messages = ""
err_message = ""
session_id: Optional[str] = None
current_role: Optional[str] = None
saw_delta = False
for line in lines:
stripped = line.strip()
if not stripped:
continue
try:
obj = json.loads(stripped)
if isinstance(obj, dict):
all_messages.append(obj)
if obj.get("session_id"):
session_id = obj.get("session_id")
current_role = update_role(obj, current_role)
is_delta = "delta" in obj or str(obj.get("type", "")).endswith("_delta")
if is_delta:
saw_delta = True
if current_role == "assistant":
if is_delta or not saw_delta:
agent_messages += extract_text(obj)
except json.JSONDecodeError:
err_message += "\n\n[json decode error] " + stripped
continue
except Exception as error:
err_message += "\n\n[unexpected error] " + f"Unexpected error: {error}. Line: {stripped!r}"
continue
return session_id, agent_messages, all_messages, err_message
def parse_json_output(lines: List[str]) -> Tuple[Optional[str], str, List[Dict[str, Any]], str]:
"""Parse json (non-stream) output into session_id, agent_messages, all_messages, error."""
err_message = ""
raw = "\n".join(lines).strip()
if not raw:
return None, "", [], "No output received from Claude Code."
try:
obj = json.loads(raw)
except json.JSONDecodeError as error:
return None, raw, [], f"Failed to parse JSON output: {error}"
session_id = None
if isinstance(obj, dict) and obj.get("session_id"):
session_id = obj.get("session_id")
agent_messages = extract_text(obj)
all_messages = [obj] if isinstance(obj, dict) else []
return session_id, agent_messages, all_messages, err_message
def main() -> None:
parser = argparse.ArgumentParser(description="Claude Code Bridge")
parser.add_argument("--PROMPT", required=True, help="Instruction for the task to send to claude.")
parser.add_argument("--cd", required=True, type=Path, help="Set the workspace root for claude before executing the task.")
session_group = parser.add_mutually_exclusive_group()
session_group.add_argument("--SESSION_ID", default="", help="Resume the specified session of claude.")
session_group.add_argument("--session-id", dest="session_id", default="", help="Use a specific session ID (UUID).")
session_group.add_argument("--continue", dest="continue_session", action="store_true", help="Continue the most recent session.")
parser.add_argument("--fork-session", action="store_true", help="Fork session when resuming/continuing.")
parser.add_argument("--no-session-persistence", action="store_true", help="Disable session persistence (print mode only).")
parser.add_argument("--model", default="", help="Model override (alias like 'sonnet'/'opus', or a full model name).")
parser.add_argument("--fallback-model", default="", help="Fallback model when the default is overloaded.")
parser.add_argument("--max-budget-usd", default="", help="Max USD budget for the call (print mode only).")
parser.add_argument("--json-schema", default="", help="JSON schema for structured output validation.")
parser.add_argument("--input-format", default="text", choices=["text", "stream-json"], help="Claude input format.")
parser.add_argument("--add-dir", action="append", default=[], help="Add additional working directories.")
parser.add_argument("--append-system-prompt", default="", help="Append text to the default system prompt.")
parser.add_argument("--system-prompt", default="", help="Replace the system prompt for the session.")
parser.add_argument("--allowed-tools", action="append", default=[], help="Tools to allow without prompting.")
parser.add_argument("--disallowed-tools", action="append", default=[], help="Tools to remove from context.")
parser.add_argument("--tools", default="", help="Comma/space-separated list of tools to enable.")
parser.add_argument("--permission-mode", default="", help="Start in a specified permission mode.")
parser.add_argument("--permission-prompt-tool", default="", help="MCP tool to handle permission prompts.")
parser.add_argument("--mcp-config", action="append", default=[], help="Load MCP servers from JSON files or strings.")
parser.add_argument("--strict-mcp-config", action="store_true", help="Only use MCP servers from --mcp-config.")
parser.add_argument("--settings", action="append", default=[], help="Load settings from JSON files or strings.")
parser.add_argument("--setting-sources", default="", help="Comma-separated list of setting sources.")
parser.add_argument("--agent", default="", help="Agent name to use for the session.")
parser.add_argument("--agents", default="", help="JSON defining custom agents.")
parser.add_argument("--output-format", default="stream-json", choices=["text", "json", "stream-json"], help="Claude output format.")
parser.add_argument("--include-partial-messages", action="store_true", help="Include partial streaming events.")
parser.add_argument("--verbose", action="store_true", help="Enable verbose CLI output (required for stream-json).")
parser.add_argument(
"--return-all-messages",
action="store_true",
help="Return all messages (e.g. tool calls, traces) from the claude session.",
)
args = parser.parse_args()
session_id_expected = args.output_format in ("json", "stream-json") and not args.no_session_persistence
cd: Path = args.cd
if not cd.exists():
result = {
"success": False,
"error": f"The workspace root directory `{cd.absolute().as_posix()}` does not exist. Please check the path and try again.",
}
print(json.dumps(result, indent=2, ensure_ascii=False))
return
cmd = ["claude", "--print", args.PROMPT, "--output-format", args.output_format, "--input-format", args.input_format]
if args.include_partial_messages:
cmd.append("--include-partial-messages")
if args.output_format == "stream-json" and not args.verbose:
cmd.append("--verbose")
elif args.verbose:
cmd.append("--verbose")
if args.model:
cmd.extend(["--model", args.model])
if args.fallback_model:
cmd.extend(["--fallback-model", args.fallback_model])
if args.max_budget_usd:
cmd.extend(["--max-budget-usd", args.max_budget_usd])
if args.json_schema:
cmd.extend(["--json-schema", args.json_schema])
if args.continue_session:
cmd.append("--continue")
if args.SESSION_ID:
cmd.extend(["--resume", args.SESSION_ID])
if args.session_id:
cmd.extend(["--session-id", args.session_id])
if args.fork_session:
cmd.append("--fork-session")
if args.no_session_persistence:
cmd.append("--no-session-persistence")
for extra_dir in args.add_dir:
cmd.extend(["--add-dir", extra_dir])
if args.system_prompt:
cmd.extend(["--system-prompt", args.system_prompt])
if args.append_system_prompt:
cmd.extend(["--append-system-prompt", args.append_system_prompt])
for tool in args.allowed_tools:
cmd.extend(["--allowedTools", tool])
if args.tools:
cmd.extend(["--tools", args.tools])
for tool in args.disallowed_tools:
cmd.extend(["--disallowedTools", tool])
if args.permission_mode:
cmd.extend(["--permission-mode", args.permission_mode])
if args.permission_prompt_tool:
cmd.extend(["--permission-prompt-tool", args.permission_prompt_tool])
for cfg in args.mcp_config:
cmd.extend(["--mcp-config", cfg])
if args.strict_mcp_config:
cmd.append("--strict-mcp-config")
for settings in args.settings:
cmd.extend(["--settings", settings])
if args.setting_sources:
cmd.extend(["--setting-sources", args.setting_sources])
if args.agent:
cmd.extend(["--agent", args.agent])
if args.agents:
cmd.extend(["--agents", args.agents])
stdout_lines, stderr_lines, returncode = read_output_lines(cmd, cwd=cd.absolute().as_posix())
if args.output_format == "stream-json":
session_id, agent_messages, all_messages, err_message = parse_stream_json(stdout_lines)
elif args.output_format == "json":
session_id, agent_messages, all_messages, err_message = parse_json_output(stdout_lines)
else:
session_id = args.SESSION_ID or args.session_id or None
agent_messages = "\n".join(stdout_lines).strip()
all_messages = []
err_message = ""
success = returncode == 0
if not success and not err_message:
err_message = f"Claude Code exited with non-zero status: {returncode}"
if session_id is None and session_id_expected:
success = False
err_message = "Failed to get `SESSION_ID` from the claude session.\n\n" + err_message
stderr_text = "\n".join(stderr_lines).strip()
if stderr_text:
err_message = (err_message + "\n\n" if err_message else "") + "[stderr]\n" + stderr_text
result: Dict[str, Any] = {"success": success}
if session_id is not None:
result["SESSION_ID"] = session_id
result["agent_messages"] = agent_messages
if success:
pass
else:
result["error"] = err_message
if args.return_all_messages:
result["all_messages"] = all_messages
print(json.dumps(result, indent=2, ensure_ascii=False))
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |