Skip to main content ホーム クリエイター adu2021 skillxiv deepcode-agentic-coding
deepcode-agentic-coding Transform research specifications into production-grade codebases through strategic information management and autonomous agent orchestration. DeepCode surpasses PhD experts and commercial tools—critical when you need scientific code reproducibility at scale.
インストールへ移動 Skills Marketplace コミュニティが作成したAIスキルを発見・探索
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/ADu2021/skillXiv --skill deepcode-agentic-codingコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Zipをダウンロード ダウンロード中... このリポジトリの他の Skills meaningful-kebab-case-name Convert arXiv papers into ready-to-use agent skills using category-aware extraction. First classifies the paper into one or more of 11 research categories, then applies a specialized extraction pipeline for each category — because different types of papers produce different types of usable knowledge. A single paper can yield multiple skills if it spans categories. Use this skill whenever the user wants to turn a paper into a skill, extract practical techniques from research, build a skill library from papers, convert arXiv papers into reusable agent instructions, or batch-process multiple papers into skills. Also trigger when someone asks about extracting actionable knowledge from papers, making research practical for LLM agents, or systematically converting academic contributions into structured agent capabilities.
action-quantization-behavior-cloning Establish regret bounds for behavior cloning with discretized actions combining statistical error and quantization error terms. Prove smoothness requirements for safe quantizer design, show that learning-based quantizers fail these requirements, and propose model-based augmentation to reduce error dependence from H² to H.
adaptive-lora-personalized-ranks Dynamically allocate LoRA ranks per-layer during fine-tuning instead of using fixed uniform ranks. Learn optimal rank for each layer and subject via variational framework with discretized exponential distribution, reducing memory footprint while maintaining fidelity and text-alignment.
name deepcode-agentic-coding title DeepCode: Open Agentic Coding version 0.0.2 engine skillxiv-v0.0.2-claude-opus-4.6 license MIT url https://arxiv.org/abs/2512.07921 keywords ["code generation","agent systems","paper-to-code","repository synthesis","agentic reasoning"] description Transform research specifications into production-grade codebases through strategic information management and autonomous agent orchestration. DeepCode surpasses PhD experts and commercial tools—critical when you need scientific code reproducibility at scale.
Overview
DeepCode treats repository synthesis as an optimization problem, strategically managing information flow to maximize relevant signals within finite context windows. The fully autonomous system orchestrates four complementary operations for code generation, transforming detailed specifications (like scientific papers) into functional, production-quality implementations.
When to Use
Converting academic papers into working code implementations
Scientific reproducibility and code generation from research
Large codebases requiring complex specification management
Scenarios needing to navigate context length limitations strategically
Applications where code quality rivals human expert work
Multi-file, complex project generation
When NOT to Use
Simple utility functions or scripts
Tasks with minimal context requirements
Projects requiring domain expertise beyond code structure
Real-time code generation with strict latency requirements
Specifications that are vague or incomplete
Core Technique
Strategic information orchestration through four complementary operations:
class DeepCodeAgent :
def __init__ (self, llm_model, context_window=8000 ):
self .model = llm_model
self .context_limit = context_window
self .operations = {
'blueprint_distillation' : BlueprintDistillation(),
'code_memory' : StatefulCodeMemory(),
'retrieval_augmented' : RetrievalAugmentedGeneration(),
'error_correction' : ClosedLoopErrorCorrection()
}
def synthesize_repository (self, specification ):
"""
Transform specification into complete codebase.
Manages information flow within context constraints.
"""
blueprint = self
self
self
self
return
def
distill_blueprint_from_specification
self, spec
"""
Compress specification to essential architecture information.
Source compression maximizes relevant signals in context.
"""
self
self
self
self
'components'
'dependencies'
'interfaces'
'algorithms'
'estimated_context'
self
self
self
'compressed_spec'
return
def
extract_components
self, specification
"""
Identify major software components from specification.
"""
f"""
Identify major software components in:
{specification[:2000 ]}
Return: component names, responsibilities, dependencies
"""
self
return
self
def
initialize_code_memory
self, blueprint
"""
Create structured memory for generated code.
Enables efficient retrieval and cross-file references.
"""
for
in
'components'
'name'
'purpose'
'dependencies'
for
in
'interfaces'
'name'
'signature'
return
def
generate_with_rag
self, blueprint, code_memory
"""
Retrieval-augmented generation strategically injects knowledge.
Retrieves relevant code context when generating new sections.
"""
for
in
'components'
'purpose'
3
self
self
'name'
'name'
return
def
construct_generation_prompt
self, component, blueprint, relevant_context, existing_code
"""
Carefully craft prompt to maximize context usage.
"""
f"""
Generate code for component: {component['name' ]}
Purpose: {component['purpose' ]}
"""
if
f"""
Reference existing implementations:
{relevant_context}
"""
f"""
Required interfaces:
{self.format_interfaces(component['dependencies' ])}
"""
if
f"""
Maintain compatibility with existing code:
{self.summarize_existing_code(existing_code)}
"""
"\n"
if
len
self
self
return
def
correct_and_refine
self, generated_files, blueprint, code_memory
"""
Closed-loop error correction refines outputs iteratively.
"""
3
for
in
range
self
if
not
break
for
in
f"""
Fix errors in {file_name} :
Current code:
{corrected_files[file_name][:1000 ]}
Errors:
{self._format_errors(error_list)}
Provide corrected code.
"""
self
return
def
test_files
self, generated_files
"""
Execute and validate generated code.
Identify errors for correction.
"""
for
in
self
if
return
def
compress_specification
self, spec, blueprint, context_limit
"""
Compress specification to fit within context window.
Keep essential information, remove redundancy.
"""
""
for
in
'algorithms'
f"\n{algo['description' ]} "
for
in
'interfaces'
f"\n{interface['signature' ]} "
if
len
return
The framework strategically optimizes information flow, achieving production-grade code quality comparable to PhD-level human experts.
Key Results
Surpasses commercial tools (Cursor, Claude Code) on code quality metrics
PhD-level expert performance on key reproduction metrics
Handles complex, multi-file repository synthesis
Outperforms prior research baselines significantly
PaperBench benchmark demonstrates effectiveness
Implementation Notes
Blueprint distillation compresses specification efficiently
Code memory enables cross-file consistency and references
RAG retrieves relevant context for each component
Closed-loop correction iteratively refines outputs
Context management crucial for large specifications
References 