Autonomous Agents Coordinating Distributed Discovery Through Emergent Artifact Exchange

Abstract

We present ScienceClaw + Infinite, a framework for autonomous scientific investigation in which independent agents conduct research without central coordination, and any contributor can deploy new agents into a shared ecosystem. The system is built around three components: an extensible registry of over 300 interoperable scientific skills, an artifact layer that preserves full computational lineage as a directed acyclic graph (DAG), and a structured platform for agent-based scientific discourse with provenance-aware governance. Agents select and chain tools based on their scientific profiles, produce immutable artifacts with typed metadata and parent lineage, and broadcast unsatisfied information needs to a shared global index. The ArtifactReactor enables plannerless coordination: peer agents discover and fulfill open needs through pressure-based scoring, while schema-overlap matching triggers multi-parent synthesis across independent analyses. An autonomous mutation layer actively prunes the expanding artifact DAG to resolve conflicting or redundant workflows, while persistent memory allows agents to continuously build upon complex epistemic states across multiple cycles. Infinite converts these outputs into auditable scientific records through structured posts, provenance views, and machine-readable discourse relations, with community feedback steering subsequent investigation cycles. Across four autonomous investigations, peptide design for the somatostatin receptor SSTR2, lightweight impact-resistant ceramic screening, cross-domain resonance bridging biology, materials, and music, and formal analogy construction between urban morphology and grain-boundary evolution, the framework demonstrates heterogeneous tool chaining, emergent convergence among independently operating agents, and traceable reasoning from raw computation to published finding.

Figures (18)

• Figure 1: ScienceClaw + Infinite ecosystem loop.(1) ScienceClaw: Agents invoke domain skills. (2) Computations: Skills execute, producing raw results. (3) Artifacts: Results wrapped as immutable, addressable records with parent lineage, accumulated in a shared DAG; need signals broadcast to a global index for peer discovery. (4) Figures: A plot agent renders visualizations from the artifact graph. (5) Infinite Interaction: Findings published as structured posts with evidence surfaces and artifact provenance. (6) Feedback: Votes, actions, and redirects from agent and human peers feed back to the ArtifactReactor's pressure scorer, which biases the next cycle toward high-impact, under-explored directions.
• Figure 2: ScienceClaw skill ecosystem. (A) Radial map of 300+ skills (as of March 13, 2026) organized into nine domain families. Bubble density within each sector reflects the relative size of that domain. (B) Skill distribution across domains, with machine learning and genomics infrastructure forming the largest categories. (C) Example agent skill chains demonstrating how different profiles compose heterogeneous analysis pipelines.
• Figure 3: Artifact workflow and cross-agent coordination. Agent A (literature seeder) produces pubmed_results artifacts and broadcasts protein_data and chemistry needs via the global index. Agent B (protein/structure) reads the index, fulfils the need with a protein_data artifact, and chains it to a sequence_alignment. Agent C (chemistry/planning) reads the index, fulfils the need with a chemistry artifact, and chains it to a retrosynthesis followed by a candidate_ranking artifacts. The ArtifactReactor performs schema-overlap matching: compatible payloads are injected as parent inputs, yielding a multi-parent synthesis artifact (parents=[a3,b2,c3]). Consumed IDs are recorded to prevent re-reaction. Only lightweight metadata (id, type, skill, agent, parents) travels to Infinite; full payloads remain in agent-local JSONL stores.
• Figure 4: ArtifactReactor Workflow: Decentralized Coordination Loop. Agents produce artifacts and broadcast unsatisfied information needs (NeedsSignal) to the global index. Peer agents scan the index for fulfillable needs (scan_needs) and compatible payloads (scan_available). Open needs are ranked by pressure score (novelty, centrality, depth, age), prioritizing convergent demand. Fulfillment proceeds via two parallel paths: need-driven reactions (run skill on demand) and schema-overlap reactions (merge compatible payloads). When multiple peer artifacts are compatible with a single skill, multi-parent synthesis merges their payloads and records all parent IDs, creating explicit cross-agent attribution. Safety mechanisms (loop prevention, self-cycle blocking, mutation layer) enforce correctness without human intervention.
• Figure 5: Infinite data model. Each post carries typed scientific fields (hypothesis, method, findings, toolsUsed) and is linked to one or more artifact records encoding the computational lineage (ID, type, skill, producerAgent, parentArtifactIds). Posts receive threaded actions and votes from agents; all activity is scoped to a community.