theseus-pro

Plugin theseus-pro from marketplace codex-marketplace in Travis-Gilbert/claude-marketplace. Commands read from the repo's manifest — 1 month ago.

Skills (1)

• intelligence-engine

  Use when tasks involve the Theseus/Index-API intelligence stack: evidence intake, claim extraction, contradiction and tension analysis, learned scoring, GNN and temporal graph memory, KGE RotatE, SBERT enrichment, GL-Fusion, model-swarm routing, compound learning, IQ measurement, self-organization,

Slash commands (7)

• /theseus-pro:architect

  System design -- two-mode deployment, feedback loop control, pipeline optimization, layer separation, domain pack architecture.

• /theseus-pro:graph

  Graph intelligence -- from objects to structure. Community detection, causal inference, self-organization, GNN embeddings, temporal memory.

• /theseus-pro:learn

  Run the compound learning pipeline. Save the session log, run Bayesian confidence updates, and present the review queue with auto-captured claims, tensions, and attention items. - Read - Edit - Write - Bash

• /theseus-pro:measure

  Intelligence measurement -- IQ tracking across 7 axes, benchmarking, trend analysis, sensitivity analysis, leverage identification.

• /theseus-pro:reason

  Epistemic reasoning -- from raw text to structured claims, tensions, and models. NER, NLI scoring, claim decomposition, confidence calibration.

• /theseus-pro:simulate

  Reasoning and simulation -- counterfactual analysis, multi-agent debate, belief revision, hypothesis generation, structural analogies.

• /theseus-pro:train

  Model training and learned intelligence -- GBT scoring, GNN embeddings, SBERT fine-tuning, LoRA LMs, evolutionary optimization, RL.

Subagents (24)

• causal-inference

  Specialist in tracing how ideas influenced each other over time. Handles temporal precedence filtering, influence DAG construction, provenance chains, and lineage tracing. Invoke when working on causal_engine.py, provenance.py, engine.py Pass 7, or any code that tracks the temporal evolution of know

• claim-analysis

  Specialist in propositional reasoning — decomposing text into atomic claims and determining support, contradiction, or neutrality between them. Handles claim decomposition (LLM and rule-based), NLI classification with CrossEncoder, stance detection, and epistemic status tracking. Invoke when working

• counterfactual-simulation

  Specialist in "what if?" reasoning over the knowledge graph. Handles dependency tree construction, counterfactual retraction cascading, fragility analysis, ATMS-style multi-context belief management, and alternative graph state comparison. Invoke when building Level 7 counterfactual simulation or an

• domain-specialization

  Specialist in adapting the engine to perform differently in different knowledge domains. Handles per-cluster model adaptation, feature importance decomposition, meta-learning for rapid domain adaptation, few-shot relation learning, schema induction, and domain pack construction. Invoke when building

• evolutionary-optimization

  Specialist in population-based search for optimal system configurations. Handles genetic algorithms, NSGA-II multi-objective optimization, CMA-ES for continuous parameters, neuroevolution, and fitness functions derived from the IQ Tracker. Invoke when tuning engine hyperparameters, optimizing multi-

• graph-neural-networks

  Specialist in learning vector representations of graph nodes by aggregating information from their neighborhoods. Handles R-GCN, CompGCN, GraphSAGE, GAT, link prediction, node classification, and structural embedding extraction. Invoke when adding GNN-based signals to the engine, training structural

• graph-theory

  Specialist in discovering structure within relationship data. Handles community detection (Louvain, Leiden), centrality measures (PageRank, betweenness, eigenvector), Personalized PageRank, structural hole analysis, spectral methods, and DAG construction. Invoke when working on graph.py, community.p

• information-retrieval

  Specialist in lexical and semantic search over growing corpora. Handles BM25 scoring, TF-IDF weighting, FAISS approximate nearest-neighbor indexes, re-ranking with cross-encoders, query expansion, and reciprocal rank fusion. Invoke when building or modifying any retrieval path — compose_engine searc

• knowledge-representation

  Specialist in structuring what is known and uncertain. Handles knowledge graph embeddings (PyKEEN RotatE, DE-SimplE), ontology design (ObjectType, ComponentType, promotion pipeline), and epistemic primitives (Claim, Question, Tension, Model, Method, Narrative). Invoke when designing data models, wor

• language-model-training

  Specialist in fine-tuning small language models on domain-specific knowledge grounded in the graph. Handles LoRA adaptation, grounded generation, task-specific training data construction, and evaluation of faithfulness and coverage. Invoke when building Level 3 hypothesis generation or any knowledge

• learned-scoring

  Specialist in replacing fixed-weight signal combination with trained models that learn which connection patterns indicate real, useful relationships versus noise. Handles feature engineering, gradient boosted trees, cold-start strategies, graceful degradation, and scorer evaluation. Invoke when work

• multi-agent-reasoning

  Specialist in adversarial epistemic debate between specialized agents. Handles the Advocate/Critic/Judge pattern, grounded debate with graph evidence, ensemble confidence scoring, and debate-as-training feedback loops. Invoke when building Level 6 multi-agent reasoning or any system where multiple A

• multimodal-networks

  Specialist in processing images, PDFs, audio, and text through unified architectures. Handles vision-language models, layout-aware document understanding, cross-modal embeddings, and continual multimodal KG growth. Invoke when extending file_ingestion.py with VLM-based extraction or building cross-m

• nlp-pipeline

  Specialist in extracting linguistic structure from natural language text. Handles NER with spaCy, adaptive entity recognition via graph-learned PhraseMatcher, sentence embeddings (SBERT, E5, Nomic), cross-encoder scoring, and tokenization strategy. Invoke when building or modifying any NLP extractio

• probabilistic-reasoning

  Manages uncertainty, confidence decay, and evidence weighting in evolving knowledge graphs. Use when working on edge strength scoring, Bayesian updates, confidence calibration, novelty dial tuning, or any system that must quantify how much it knows vs. how much it guesses. Examples: - <example>User

• program-synthesis

  Turns stored knowledge into executable procedures via a domain-specific language. Use when working on Method definitions, DSL design, rule compilation, method versioning, or any system that bridges captured information and encoded executable knowledge. Examples: - <example>User asks "how do I turn t

• reinforcement-learning

  Specialist in learning policies from interaction rewards. Handles contextual bandits for engine configuration selection, Thompson sampling, MINERVA-style multi-hop path reasoning, and reward shaping from user engagement. Invoke when building Level 5 self-modifying pipeline or any system that learns

• self-organization

  Designs and maintains feedback loops that let the knowledge graph modify its own structure safely. Use when working on auto-classification, community clustering, entity promotion, edge decay, emergent type detection, or any system that reorganizes itself based on its own outputs. Examples: - <exampl

• software-architecture

  Makes epistemic systems run reliably in production across constrained and rich runtimes. Use when working on deployment, the two-mode contract, RQ task queues, caching layers, memory budgets, graceful degradation, Modal GPU dispatch, or any infrastructure concern that affects how the system operates

• symbolic-reasoning

  Specialist in logic-based inference over graph structures. Handles defeasible logic (PyReason), AGM belief revision, Truth Maintenance Systems, rule learning from KG structure (RNNLogic), and structure- mapping analogy (ANASIME). Invoke when building Level 4 emergent ontology rules, Level 7 TMS infr

• systems-theory

  Meta-discipline specialist for reasoning about Theseus as a complex adaptive system. Handles feedback loop classification, attractor dynamics, sensitivity analysis, gradual loop closure, phase transitions, information-theoretic capacity, and separation of concerns in layered systems. Invoke when ana

• temporal-graph-memory

  Specialist in modeling how graphs evolve over time. Handles Temporal Graph Networks (TGN), event-based node memory, DE-SimplE temporal embeddings, and sliding-window graph dynamics. Invoke when working on temporal_memory.py, temporal_evolution.py, or any code that tracks how knowledge and connection

• training-pipeline

  Improves retrieval, classification, and knowledge graph models from accumulated user data. Use when working on triplet construction, fine-tuning SBERT, active learning, KGE training, evaluation metrics, or any system that learns from the knowledge graph's own data. Examples: - <example>User asks "ho

• web-acquisition

  Gets evidence into the epistemic system from the web. Use when working on content extraction, URL capture, PDF ingestion, Firecrawl pipelines, source typing, corpus construction, or any system that brings external information into the knowledge graph. Examples: - <example>User asks "how do we scrape