SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2304.07590 · 2023

Self-collaboration Code Generation via ChatGPT

arXiv · PDF · Open in the Atlas

Code that ran

11 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 11 have no contract verdict; 0 failures of our instrument; 6 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2304.07590.json · badge: /api/ran/2304.07590.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
YihongDong/Self-collaboration-Code-Generation canonical (classifier label)
classifier basis: census
11 ran · 6 no recorded run
FunctionStatusWhere it lives
code_truncate Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/utils.py
code served (permissive licence) · get_code("39ca69f20203b0e1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def code_truncate_regex(code):
    code_regex = '```(.*?|)\\n(?P<code>.*?)```'
    match = re.search(code_regex, code, re.DOTALL)
    code = match.group('code') if match else ''
    return code

def code_truncate(response):
    code = code_truncate_regex(response)
    if code == '':
        generation = response[response.find('def'):]
        tem = [s for s in generation.split('\n\n') if 'def ' in s or s[:1] == ' ']
        code = '\n\n'.join(tem).strip('```').strip()
    return code

# Test case 1: response with markdown code block
response_with_markdown = """```python
def hello():
    print("world")
"""

# Test case 2: response without markdown code block but with def
response_with_def = """Some explanation here
def add(a, b):
    return a + b

def multiply(a, b):
    return a * b"""

# Test case 3: response with code block and language specified
response_with_lang = """Here's the code:
javascript
function test() {
    console.log("test");
}
"""

ARGS = {
    'response': response_with_markdown
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
code_truncate_regex Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/utils.py
code served (permissive licence) · get_code("4d6900765a680c9f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def code_truncate_regex(code):
    code_regex = '```(.*?|)\\n(?P<code>.*?)```'
    match = re.search(code_regex, code, re.DOTALL)
    code = match.group('code') if match else ''
    return code

# Test case 1: code with markdown code block
test_code_1 = """```python
def hello():
    print("world")
"""

# Test case 2: code without markdown code block
test_code_2 = "just some plain text without code block"

# Test case 3: code with empty code block
test_code_3 = """
"""

# Test case 4: code with language specified
test_code_4 = """```javascript
console.log("test");
"""

ARGS = {
    "code": test_code_1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_patch Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/patch_utils.py
code served (permissive licence) · get_code("40263bfc7c17c517")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

# SETUP CODE
ARGS = {
    "response": """Here's a unified diff patch:
diff
--- a/src/file.py
+++ b/src/file.py
@@ -10,5 +10,6 @@
 def hello():
     print("world")
-    x = 1
+    x = 2
+    y = 3
 return x

"""
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_image_name Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/run_swe_docker.py
code served (permissive licence) · get_code("8d702cbfc56a7d48")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import subprocess
import platform

# Mock the docker info command to avoid actual Docker dependency
def mock_docker_info(*args, **kwargs):
    class MockResult:
        returncode = 0
        stdout = "x86_64"
        stderr = ""
    return MockResult()

# Patch subprocess.run for this test
original_run = subprocess.run
subprocess.run = mock_docker_info

# Now import and set up the function
def _detect_arch():
    """Detect architecture from Docker daemon, not from current process."""
    try:
        result = subprocess.run(['docker', 'info', '--format', '{{.Architecture}}'], capture_output=True, text=True, timeout=5)
        if result.returncode != 0:
            raise RuntimeError(f'docker info failed: {result.stderr.strip()}')
        arch = result.stdout.strip()
        if not arch:
            raise RuntimeError('docker info returned empty architecture')
        if 'arm' in arch or 'aarch' in arch:
            return 'arm64'
        return 'x86_64'
    except Exception as e:
        fallback = 'arm64' if platform.machine() in ('arm64', 'aarch64') else 'x86_64'
        print(f'  WARNING: arch detection failed ({e}), falling back to {fallback}')
        return fallback

_cfg = {'arch': _detect_arch()}

def get_image_name(instance_id):
    return f"ghcr.io/epoch-research/swe-bench.eval.{_cfg['arch']}.{instance_id}:latest"

# Restore original subprocess.run
subprocess.run = original_run

# Define arguments for get_image_name function
ARGS = {
    'instance_id': 'test-instance-12345'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_repo_structure Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/repo_tools.py
code served (permissive licence) · get_code("34c188c42a3e46a9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
import shutil

# Create a temporary directory structure to test the function
temp_dir = tempfile.mkdtemp()

# Create some subdirectories and files
os.makedirs(os.path.join(temp_dir, "src"), exist_ok=True)
os.makedirs(os.path.join(temp_dir, "src", "models"), exist_ok=True)
os.makedirs(os.path.join(temp_dir, "tests"), exist_ok=True)
os.makedirs(os.path.join(temp_dir, ".hidden"), exist_ok=True)
os.makedirs(os.path.join(temp_dir, "__pycache__"), exist_ok=True)

# Create some files
open(os.path.join(temp_dir, "README.md"), "w").close()
open(os.path.join(temp_dir, "setup.py"), "w").close()
open(os.path.join(temp_dir, "src", "main.py"), "w").close()
open(os.path.join(temp_dir, "src", "models", "network.py"), "w").close()
open(os.path.join(temp_dir, "tests", "test_main.py"), "w").close()
open(os.path.join(temp_dir, ".hidden", "secret.txt"), "w").close()

ARGS = {
    "repo_path": temp_dir,
    "max_depth": 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
openai_config Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/config.py
code served (permissive licence) · get_code("ba1bea1b924882e1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ModelConfig. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
from dataclasses import dataclass, field

_DEFAULT_MODEL = os.environ.get('MODEL', 'moonshotai/kimi-k2.5')
_DEFAULT_BASE_URL = os.environ.get('BASE_URL', 'https://openrouter.ai/api/v1')
_DEFAULT_API_KEY_ENV = os.environ.get('API_KEY_ENV', 'OPENROUTER_API_KEY')

@dataclass
class ModelConfig:
    model: str = ''
    base_url: str = ''
    api_key_env: str = ''
    max_tokens: int = 4096
    temperature: float = 0.0
    top_p: float = 0.95

    def __post_init__(self):
        if not self.model:
            self.model = _DEFAULT_MODEL
        if not self.base_url:
            self.base_url = _DEFAULT_BASE_URL
        if not self.api_key_env:
            self.api_key_env = _DEFAULT_API_KEY_ENV

def openai_config(model='gpt-3.5-turbo-0301', max_tokens=512, temperature=0.0, top_p=0.95):
    """Create a ModelConfig compatible with the original OpenAI mode."""
    return ModelConfig(model=model, base_url='https://api.openai.com/v1', api_key_env='OPENAI_API_KEY', max_tokens=max_tokens, temperature=temperature, top_p=top_p)

ARGS = {
    'model': 'gpt-3.5-turbo-0301',
    'max_tokens': 512,
    'temperature': 0.0,
    'top_p': 0.95
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ModelConfig — a ModelConfig output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ModelConfig output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
prompt_split_humaneval Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/utils.py
code served (permissive licence) · get_code("91120a9f594cb6e3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

# Create a sample prompt string similar to what HumanEval datasets contain
prompt = """
def add(a, b):
    \"\"\"Add two numbers together.
    
    Example:
    >>> add(1, 2)
    3
    >>> add(5, 10)
    15
    \"\"\"
    return a + b
"""

method_name = "add"

ARGS = {
    "prompt": prompt,
    "mehotd_name": method_name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_file Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/repo_tools.py
code served (permissive licence) · get_code("bac44134ee8d1581")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()
repo_path = temp_dir

# Create a test file within the repo
test_file_path = os.path.join(repo_path, "test.txt")
with open(test_file_path, 'w') as f:
    f.write("Line 1\n")
    f.write("Line 2\n")
    f.write("Line 3\n")

# Define arguments for read_file function
ARGS = {
    'repo_path': repo_path,
    'file_path': 'test.txt'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
repair_patch Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/patch_utils.py
code served (permissive licence) · get_code("5b0f4aae224d17d9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re
from typing import List, Tuple

# Construct a valid patch string with common truncation issues
patch_content = """--- a/file.txt
+++ b/file.txt
@@ -1,5 +1,6 @@
 line 1
 line 2
-line 3
+line 3 modified
 line 4
 line 5
+new line"""

ARGS = {
    "patch": patch_content
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
strip_context_lines Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/patch_utils.py
code served (permissive licence) · get_code("4699f7376865432a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re
from typing import List, Tuple

# Construct a valid patch string with headers, context lines, and +/- lines
patch_content = """--- a/file.py
+++ b/file.py
@@ -10,6 +10,8 @@
 context line 1
 context line 2
-removed line
+added line
 context line 3
 context line 4
"""

ARGS = {
    "patch": patch_content,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
unsafe_execute Ran
ran on a synthesized input
YihongDong/Self-collaboration-Code-Generation/core/sandbox.py
code served (permissive licence) · get_code("584acc1f9637179b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# The function `unsafe_execute` takes two parameters:
# 1. code: a string containing Python code to execute
# 2. report: a string to append to the code before execution

# Simple test code that will pass
code = """
x = 1 + 1
assert x == 2, "Basic math failed"
"""

# Simple report string
report = """
print("Test completed")
"""

ARGS = {
    "code": code,
    "report": report,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
call_chatgpt Not yet run
unverified (harvested, no recorded run)
YihongDong/Self-collaboration-Code-Generation/core/backend.py
code served (permissive licence) · get_code("edb3468adda5e752")
How it ran

no invocation recorded.

call_llm Not yet run
unverified (harvested, no recorded run)
YihongDong/Self-collaboration-Code-Generation/core/backend.py
code served (permissive licence) · get_code("6b210bdeeae33ffc")
How it ran

no invocation recorded.

call_llm_with_tools Not yet run
unverified (harvested, no recorded run)
YihongDong/Self-collaboration-Code-Generation/core/backend.py
code served (permissive licence) · get_code("93b3e7a2b08c74b7")
How it ran

no invocation recorded.

clone_repo Not yet run
unverified (harvested, no recorded run)
YihongDong/Self-collaboration-Code-Generation/core/repo_tools.py
code served (permissive licence) · get_code("85dcee64dcb03ae5")
How it ran

no invocation recorded.

image_exists Not yet run
unverified (harvested, no recorded run)
YihongDong/Self-collaboration-Code-Generation/run_swe_docker.py
code served (permissive licence) · get_code("6d314c06f83ec007")
How it ran

no invocation recorded.

start_container Not yet run
unverified (harvested, no recorded run)
YihongDong/Self-collaboration-Code-Generation/run_swe_docker.py
code served (permissive licence) · get_code("c94bc015446c731b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 17 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Although Large Language Models (LLMs) have demonstrated remarkable code-generation ability, they still struggle with complex tasks. In real-world software development, humans usually tackle complex tasks through collaborative teamwork, a strategy that significantly controls development complexity and enhances software quality. Inspired by this, we present a self-collaboration framework for code generation employing LLMs, exemplified by ChatGPT. Specifically, through role instructions, 1) Multiple LLM agents act as distinct `experts', each responsible for a specific subtask within a complex task; 2) Specify the way to collaborate and interact, so that different roles form a virtual team to facilitate each other's work, ultimately the virtual team addresses code generation tasks collaboratively without the need for human intervention. To effectively organize and manage this virtual team, we incorporate software-development methodology into the framework. Thus, we assemble an elementary team consisting of three LLM roles (i.e., analyst, coder, and tester) responsible for software development's analysis, coding, and testing stages. We conduct comprehensive experiments on various code-generation benchmarks. Experimental results indicate that self-collaboration code generation relatively improves 29.9%-47.1% Pass@1 compared to the base LLM agent. Moreover, we showcase that self-collaboration could potentially enable LLMs to efficiently handle complex repository-level tasks that are not readily solved by the single LLM agent.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2304.07590")
get_code_for_paper("2304.07590")
have("2304.07590")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2304.07590.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2304.07590.svg)](https://syntology.ai/paper/2304.07590)

Connect an agent — have() is free.