DevOps & Infra

ndpvt-web Skill

Build agentic image retrieval systems that perform multi-step contextual reasoning over visual histories instead of isolated semantic matching. Use when: 'build a context-aware image search agent', 'retrieve images using temporal reasoning', 'search photos by contextual clues across events', 'implement dual-memory agent for image retrieval', 'create a visual history exploration pipeline', 'benchmark multimodal agents on retrieval tasks'.

ndpvt-web Skill

Build multi-agent emotion-focused therapy (EFT) reasoning pipelines for empathetic mental health Q&A systems. Uses a bottom-up three-stage chain-of-thought: Embodied Perception, Cognitive Exploration, and Narrative Intervention with eight specialized agents. Trigger phrases: 'build an EFT chatbot', 'emotion-focused therapy agent', 'empathetic counseling system', 'multi-agent mental health pipeline', 'somatic-aware therapy bot', 'EFT-CoT reasoning chain'.

ndpvt-web Skill

Apply the REFLECT constitutional alignment framework to enforce user-defined principles on LLM outputs through a multi-stage pipeline: constitution-conditioned generation, self-evaluation with Likert scoring, self-critique, and principled revision. Use this skill when asked to: "align output to these principles", "apply constitutional review to this response", "reflect on whether this follows our guidelines", "enforce these coding standards on generated code", "review this output against our policy", "run a principled critique and revision pass".

ndpvt-web Skill

Implement multi-agent RAG systems with coordinated sequential-parallel scaling and shared context management for complex multi-hop question answering. Use when: 'build a multi-hop RAG pipeline', 'implement adaptive retrieval with parallel branching', 'create a RAG system that scales retrieval depth and width', 'build a multi-agent retrieval framework', 'implement SPARC-RAG pattern for complex QA', 'design a RAG system with context management across retrieval rounds'.

ndpvt-web Skill

Diagnose and repair failing computational research code to restore reproducibility. Uses an agent-based iterative workflow: inspect files, identify failures (missing packages, broken paths, version conflicts, missing logic), apply targeted fixes, and rerun in isolated environments. Trigger phrases: 'reproduce this analysis', 'fix this R script', 'make this code reproducible', 'debug this research pipeline', 'repair computational workflow', 'rerun this study'

ndpvt-web Skill

Translate natural language optimization problems into executable solver code using a Canonical Intermediate Representation (CIR) schema and multi-agent R2C pipeline. Decomposes operational rules into constraint archetypes and modeling paradigms before generating code. Triggers: "formulate this optimization problem", "write a solver for this scheduling problem", "convert these business rules to constraints", "model this linear program from the description", "generate Gurobi/PuLP code for this OR problem", "help me formulate these operational constraints mathematically".

ndpvt-web Skill

Build and evaluate long-term conversational memory systems for multi-party, multi-topic dialogues. Implements the EverMemBench framework for stress-testing memory architectures against realistic workplace conversation patterns with temporal evolution, cross-topic interleaving, and role-specific personas. Use when: 'build a memory system for multi-user chat', 'evaluate my RAG memory pipeline', 'benchmark long-term conversation recall', 'test memory across multi-party dialogues', 'design a temporal memory store for chat agents', 'audit retrieval quality for conversational AI'.

ndpvt-web Skill

Detect, confirm, repair, and validate recurring software vulnerabilities using a multi-agent pipeline modeled on MAVM. Builds a vulnerability knowledge base from historical CVE/patch data, then coordinates specialized agents for detection, confirmation, repair, and validation. Trigger phrases: 'find recurring vulnerabilities', 'scan for known vulnerability patterns', 'port security patches across repos', 'detect unpatched code clones', 'end-to-end vulnerability management', 'fix recurring security issues'.

ndpvt-web Skill

Evaluate and harden LLM-based autonomous agents against adversarial attacks using the α³-SecBench layered security framework. Assesses security (attack detection, CWE attribution), resilience (safe degradation), and trust (policy-compliant tool usage) across 7 autonomy layers. Use when: 'audit my LLM agent for security', 'add adversarial resilience to my autonomous system', 'evaluate agent trust and tool safety', 'harden my AI agent against prompt injection', 'security benchmark my LLM pipeline', 'test my agent for hallucinated tool calls'.

ndpvt-web Skill

Build production retrieval systems using pplx-embed, diffusion-pretrained dense and contextualized embedding models with INT8 quantization, late chunking for long documents, and multi-stage contrastive training. Use when: 'build a semantic search pipeline', 'set up document retrieval with contextual embeddings', 'implement late chunking for long documents', 'create a multilingual search index', 'optimize embedding storage with quantization', 'add contextualized passage retrieval to RAG'.

ndpvt-web Skill

Iteratively optimize LLM prompts for information extraction tasks using self-evaluation feedback loops. Applies the KRPO framework: extract structured data, restore it to natural language, score semantic consistency via NLI, then generate textual gradients to refine the prompt. Includes relation canonicalization to deduplicate and normalize extracted schemas. Trigger phrases: - "Extract relations from text and optimize the prompt" - "Build a self-improving extraction pipeline" - "Optimize my prompt for triplet extraction" - "Extract knowledge graph triples from unstructured text" - "Set up iterative prompt refinement with feedback" - "Canonicalize extracted relations across documents"

ndpvt-web Skill

Build multidimensional profiling pipelines for large scientific paper corpora. Combines BERTopic clustering, LLM-structured extraction, and weighted semantic retrieval to analyze research trends, topic lifecycles, dataset/model adoption, and methodological shifts. Use when: 'profile these research papers', 'analyze trends in this paper corpus', 'build a scientific literature pipeline', 'extract structured knowledge from papers', 'track topic evolution across conferences', 'find emerging research directions'.

ndpvt-web Skill

Build and evaluate biomedical RAG pipelines using the MRAG benchmark methodology. Configures retrieval, prompting, and generation components for medical QA systems. Use when: 'build a medical RAG pipeline', 'evaluate my biomedical QA system', 'optimize retrieval for clinical questions', 'set up PubMed-based RAG', 'benchmark RAG on medical datasets', 'configure RAG for drug interaction extraction'.

ndpvt-web Skill

Build production-grade natural language to SQL/API pipelines using RAG variant selection (Standard RAG, Self-RAG, CoRAG). Implements iterative query decomposition, hybrid documentation retrieval, and dynamic task classification for enterprise NL interfaces. Trigger phrases: - "Build a natural language to SQL interface with RAG" - "Generate API calls from user questions using retrieval augmented generation" - "Set up a hybrid SQL and REST API generation pipeline" - "Implement CoRAG for enterprise query generation" - "Create a text-to-SQL system with document retrieval" - "Design a retrieval pipeline that handles both database queries and API calls"

ndpvt-web Skill

Implement the Fat-Cat document-driven metacognitive agent architecture for complex multi-step reasoning tasks. Uses Markdown documents as global state instead of JSON, a four-stage reasoning pipeline (metacognitive analysis, strategy selection, step decomposition, execution), textual strategy evolution for accumulating task-solving knowledge, and a closed-loop watcher to prevent hallucinations and infinite loops. Trigger phrases: "use fat-cat for this task", "document-driven agent", "metacognitive reasoning pipeline", "markdown state management", "multi-agent with strategy evolution", "fat-cat agent workflow"

ndpvt-web Skill

LLM-driven precision querying of structured tabular data via Python/Pandas code generation and retrieval-augmented extraction from unstructured clinical text. Use when: 'query this table in natural language', 'extract information from clinical notes', 'build a RAG pipeline for medical records', 'generate Pandas code from a question', 'answer questions about EHR data', 'set up evaluation for Q&A over datasets'.

ndpvt-web Skill

Build contamination-free LLM evaluation pipelines with multi-agent data curation and automated rubric-based scoring. Uses LiveMedBench's three-agent curation framework and bipolar rubric evaluation to assess LLM outputs against granular, case-specific criteria. Trigger phrases: 'build a contamination-free benchmark', 'evaluate LLM with rubrics', 'curate clinical test data', 'automated rubric evaluation pipeline', 'detect data contamination in LLMs', 'multi-agent data curation framework'.

ndpvt-web Skill

Compile domain ontologies (OWL/RDFS/JSON-LD schemas) into executable tool interfaces with embedded semantic constraints, so LLM agents enforce domain rules during generation rather than post-hoc. Use when: 'compile my ontology into tools', 'enforce schema constraints in agent tools', 'generate MCP tools from OWL', 'build knowledge graph extraction pipeline', 'ontology-driven tool generation', 'semantic constraint enforcement for agents'.

ndpvt-web Skill

Apply Theory-of-Mind (ToM) guided reasoning chains to multimodal emotion analysis tasks. Decomposes emotional reasoning into hierarchical cognitive levels—perception, understanding, and causal cognition—tracking mental states explicitly before reaching conclusions. Use when: 'analyze the emotions in this image/video', 'why does this person feel that way', 'build an emotion reasoning pipeline', 'detect sarcasm or humor in multimodal content', 'evaluate emotional understanding in my model', 'add ToM-based reasoning to my MLLM'.

ndpvt-web Skill

Implement available-but-invisible privacy protection for mobile GUI agents using PII-aware anonymization with deterministic, type-preserving placeholders. Use when: 'anonymize PII in UI automation', 'build privacy layer for mobile agent', 'protect sensitive data in screenshots', 'add PII detection to Android agent', 'type-preserving placeholder system', 'privacy-safe GUI agent pipeline'.

ndpvt-web Skill

Build and optimize two-stage RAG pipelines with bi-encoder retrieval, cross-encoder re-ranking, and empirically-validated chunking strategies. Use when: 'build a RAG pipeline', 'add re-ranking to retrieval', 'optimize chunking for documents', 'set up document QA with re-ranking', 'improve RAG faithfulness', 'two-stage retrieval pipeline'.

ndpvt-web Skill

Build multi-label hierarchical classifiers for MITRE ATT&CK text tagging using stage-wise classical ML (SGD-SVM + TF-IDF). Use when: 'tag CTI text with ATT&CK', 'classify threat reports with MITRE tactics', 'build hierarchical cybersecurity classifier', 'map CVE descriptions to ATT&CK techniques', 'automate MITRE tagging pipeline', 'multi-label threat classification'.

ndpvt-web Skill

Build event-driven video stream processing pipelines that detect meaningful state transitions instead of processing every frame. Use when asked to: 'build a real-time video understanding system', 'detect events in a video stream', 'process long video with memory', 'reduce redundant frame processing', 'stream video to LLM efficiently', 'build an event-aware video pipeline'.

ndpvt-web Skill

Build self-evolving multi-agent systems that accumulate tool-level expertise through structured interaction without model fine-tuning. Uses GeoEvolver's architecture: retrieval-augmented orchestration, parallel sub-goal exploration, contrastive memory distillation, and root-cause failure attribution. Triggers: 'build a self-evolving agent pipeline', 'create an experience-driven multi-agent system', 'add memory to my agent workflow', 'implement tool exploration with failure learning', 'make agents learn from execution history', 'build a GeoEvolver-style system'

ndpvt-web Skill

Build iterative RAG pipelines that construct evolving knowledge graphs to anchor retrieval across multiple hops. Use when user says 'multi-hop QA', 'graph-guided retrieval', 'iterative RAG', 'knowledge graph indexing', 'connect evidence across documents', or 'structured retrieval pipeline'.

ndpvt-web Skill

Build hybrid classifier-then-LLM pipelines to extract actionable suggestions from unstructured customer reviews. Use when the user says 'extract suggestions from reviews', 'mine actionable feedback', 'analyze customer complaints for improvements', 'build a suggestion extraction pipeline', 'classify and cluster review feedback', or 'summarize actionable insights from user reviews'.

ndpvt-web Skill

Build and evaluate LLM-based Instructional Design agents using the ADDIE framework, Context Matrix scenario generation, and multi-judge evaluation. Triggers: 'design a course using ADDIE', 'build an instructional design agent', 'evaluate my ISD pipeline', 'create training program with learning objectives', 'benchmark educational content generation', 'generate instructional scenarios with context variables'

ndpvt-web Skill

Apply Foundation-Sec-8B-Reasoning cybersecurity reasoning patterns: structured <think> chain-of-thought for CVE-to-CWE mapping, MITRE ATT&CK classification, CVSS scoring, threat intelligence analysis, and multi-hop vulnerability reasoning. Use when the user asks to "analyze a CVE", "map vulnerabilities to CWE", "classify attack techniques", "reason about security threats", "triage a vulnerability", or "build a cybersecurity reasoning pipeline".

ndpvt-web Skill

Implement the ExperienceWeaver hierarchical experience-learning framework to improve text quality from small feedback sets. Distills noisy corrections into structured Tips and Strategies, then injects them into a multi-agent detection-revision-critique pipeline. Use when: 'improve text from few examples', 'learn revision patterns from feedback', 'build experience-based text correction', 'distill feedback into reusable rules', 'small-sample text improvement pipeline', 'create agentic revision system from examples'.

ndpvt-web Skill

Decouple reasoning from embedding compression in multimodal retrieval pipelines by enriching queries and corpus entries with explicit semantic context before encoding. Use when: 'build a multimodal search system', 'improve image-text retrieval accuracy', 'fix retrieval for ambiguous queries', 'add reasoning to my embedding pipeline', 'my CLIP search returns wrong results for complex queries', 'enhance retrieval with VLM captions'.

ndpvt-web Skill

Decompose hate speech detection into a checklist of ten concept-level binary questions answered independently by an LLM, then aggregate results via a lightweight decision tree for interpretable, cross-dataset-robust classification. Use when asked to: 'build a hate speech detector', 'create an interpretable content moderation system', 'detect hateful content with explainability', 'classify toxic text with audit trails', 'implement checklist-based text classification', 'make a robust hate speech pipeline'.

revfactory Bundle

An exam preparation full pipeline. An agent team collaborates to perform trend analysis, weakness diagnosis, customized study planning, mock exam creation, and error analysis. Use this skill for requests like 'help me prepare for an exam', 'create a mock exam', 'analyze past exams', 'diagnose weaknesses', 'error analysis', 'college entrance prep', 'certification exam', 'civil service exam', 'TOEIC prep', 'create a study plan', and other exam preparation needs. Existing score reports or error data can be used to augment the diagnostic phase. However, actual exam registration/payment, academy recommendations, and live lecture delivery are outside the scope of this skill.

revfactory Bundle

A pipeline where an agent team systematically creates Architecture Decision Records (ADRs). Use this skill for requests such as 'write an ADR,' 'architecture decision record,' 'document a technical decision,' 'architecture decision record,' 'organize architecture selection rationale,' 'technology stack decision,' 'alternative comparison analysis,' 'tradeoff analysis,' or 'architecture decision history.' Note: actual code migration execution, infrastructure provisioning, and performance test execution are outside the scope of this skill.

ffsshhttiikk Skill

ETL pipeline design and implementation

ffsshhttiikk Skill

Cloud security best practices and implementation

ffsshhttiikk Skill

Open-source automation platform for configuration management, application deployment, and IT orchestration