name: clawmind version: 2.0.0 kind: agentic-infrastructure track: "Track 1: Agentic Infrastructure & OpenClaw Lab" description: > Multi-agent Web3 due diligence powered by 0G Compute for agent inference, 0G Storage for reports and memory, and 0G Chain for report anchoring. runtime: framework: nextjs language: typescript entrypoints: api: analyze: app/api/analyze/route.ts memory: app/api/memory/route.ts retrieve_report: app/api/report/retrieve/route.ts orchestrator: lib/orchestrator/run-analysis.ts orchestration: mode: sequential-multi-agent-pipeline manifest_driven: true state_persistence: primary: 0G_STORAGE fallback: LOCAL_FALLBACK compute: primary: 0G_COMPUTE_ROUTER fallback: LOCAL_DETERMINISTIC_INFERENCE pipeline: - id: memory_retrieval label: Memory Retrieval skill: persistent-memory-retrieval model: local temperature: 0 max_tokens: 0 input: user_task output: relevant_memories depends_on: [] reads: - local_memory_store - zero_g_memory_index - id: planner label: Planner Agent skill: task-decomposition model: deepseek/deepseek-chat-v3-0324 temperature: 0.2 max_tokens: 1200 structured_output: true input: - user_task - relevant_memories output: execution_plan depends_on: - memory_retrieval - id: researcher label: Research Agent skill: research-extraction model: deepseek/deepseek-chat-v3-0324 temperature: 0.3 max_tokens: 1500 structured_output: true input: - user_task - execution_plan output: research_findings depends_on: - planner - id: risk_agent label: Risk Agent skill: web3-risk-analysis model: deepseek/deepseek-chat-v3-0324 temperature: 0.2 max_tokens: 1500 structured_output: true input: - user_task - research_findings - relevant_memories output: risk_map depends_on: - researcher - memory_retrieval - id: architect label: Architect Agent skill: architecture-design model: deepseek/deepseek-chat-v3-0324 temperature: 0.3 max_tokens: 1500 structured_output: true input: - user_task - research_findings - risk_map output: architecture_recommendations depends_on: - researcher - risk_agent - id: critic label: Critic Agent (Adversarial) skill: adversarial-review model: deepseek/deepseek-chat-v3-0324 temperature: 0.8 max_tokens: 1500 structured_output: true input: - execution_plan - research_findings - risk_map - architecture_recommendations output: critique depends_on: - planner - researcher - risk_agent - architect - id: final_agent label: Final Decision Agent skill: decision-synthesis model: deepseek/deepseek-chat-v3-0324 temperature: 0.1 max_tokens: 2500 structured_output: true input: - user_task - relevant_memories - execution_plan - research_findings - risk_map - architecture_recommendations - critique output: structured_decision_report depends_on: - planner - researcher - risk_agent - architect - critic - id: memory_writer label: Memory Writer skill: persistent-memory-writing model: all-MiniLM-L6-v2 + 0G Storage temperature: 0.1 max_tokens: 500 structured_output: true input: - structured_decision_report - storage_receipt output: - memory_record - zero_g_memory_index_receipt - on_chain_receipt depends_on: - final_agent skills: - id: persistent-memory-retrieval description: Retrieves prior ClawMind memory records using embedding-based semantic similarity and ranks them against the new task. embedding_model: all-MiniLM-L6-v2 retrieval_method: cosine_similarity_top_k - id: task-decomposition description: Breaks ambiguous Web3/AI tasks into analysis work items. model_family: deepseek - id: research-extraction description: Extracts project claims, assumptions, stakeholders, and missing information. model_family: deepseek - id: web3-risk-analysis description: Identifies custody, oracle, governance, tokenomics, security, and automation risks. model_family: deepseek - id: architecture-design description: Proposes agent, infra, storage, and execution-control architecture. model_family: deepseek - id: adversarial-review description: Hostile adversarial review. Runs as a separate high-temperature Critic role on the production 0G Compute route. Challenges assumptions, identifies blind spots, and flags optimism bias. model_family: deepseek - id: decision-synthesis description: Converts agent outputs into a structured score, recommendation, risks, opportunities, and next steps. model_family: deepseek - id: persistent-memory-writing description: Stores distilled analysis memory with embeddings and persists the memory index to 0G Storage. model_family: local_embeddings artifacts: reports: kind: CLAWMIND_ANALYSIS_REPORT storage: 0G_STORAGE uri_scheme: 0g:// memory_index: kind: CLAWMIND_MEMORY_INDEX storage: 0G_STORAGE uri_scheme: 0g:// embedding_model: all-MiniLM-L6-v2 embedding_dimensions: 384 retrieval: cosine_similarity_top_k security: execution_policy: - LLM agents may reason and recommend. - LLM agents must not directly sign transactions. - Human or deterministic policy layer must gate fund movement. wallet_policy: - Use burner wallets with limited 0G token balance for mainnet operations. - Never commit private keys. fallback_policy: - Fallback mode must be clearly labeled as LOCAL_FALLBACK in receipts. zero_g_integration: compute: provider: 0G_COMPUTE_ROUTER endpoint: https://router-api.0g.ai/v1/chat/completions models: - id: deepseek/deepseek-chat-v3-0324 roles: [planner, researcher, risk_agent, architect, critic, final_agent] family: deepseek strategy: single_primary_model_route strategy_description: > Production inference currently routes all LLM agents through the stable DeepSeek route on 0G Compute. Agent diversity comes from separate roles, role-specific prompts, Critic temperature, and explicit score-adjustment math. Additional 0G Compute models remain configurable in the runtime model router when availability allows. storage: provider: 0G_STORAGE indexer_rpc: https://indexer-storage-turbo.0g.ai evm_rpc: https://evmrpc.0g.ai uri_scheme: 0g:// description: Reports and memory indexes are persisted to 0G Storage with verifiable receipts. chain: network: mainnet chain_id: 16661 contract: AnalysisRegistry.sol contract_address: "0x08a9c275f5d0764a32f9dda4f50ba6f9a828e2b1" explorer: https://chainscan.0g.ai/address/0x08a9c275f5d0764a32f9dda4f50ba6f9a828e2b1 description: Each completed analysis is anchored on-chain with root hash, score, recommendation, storage URI, and an EIP-712 signature from an authorized ClawMind operator. authentication: EIP712_OPERATOR_SIGNATURE judge_mode: url: /judge api: /api/judge description: Read-only review surface for hackathon judges. Shows 0G integration evidence, on-chain analysis data, and memory stats.