DevOps & Infra

tuyv Bundle

Configures best-practice alerting policies for AI agents using OpenTelemetry (OTel) metrics, generating output as Terraform (.tf) configuration files. Use when analyzing, writing, or deploying alerting policies to monitor agent latency, error rates, token usage, and quality metrics. Don't use for standard infrastructure monitoring unrelated to AI agents, or when the agent is not instrumented with OpenTelemetry (for Reliability, Cost, Safety, Security alerts). NOTE: Reliability, Cost, Safety, and Security alerts use generic OTel metrics and work across runtimes (such as Cloud Run, Vertex AI). Quality alerts rely on Vertex AI Online Monitors and are strictly bound to Vertex AI deployments.

tuyv Bundle

Diagnose failed Spark jobs, unhealthy Livy sessions, and performance bottlenecks in Microsoft Fabric via read-only CLI triage. Use ONLY for FAILED/unhealthy runs; running a notebook and reporting its success status is `spark-authoring-cli`. Use when the user wants to: (1) diagnose why a Spark job, notebook run, or Lakehouse job failed, (2) triage stuck or dead Livy sessions, (3) identify OOM, shuffle spill, or data skew, (4) retrieve driver/executor logs or Spark Advisor findings, (5) copy event logs and start a local Spark History Server, (6) diagnose Spark activities in a failed pipeline run. Triggers: "diagnose my failed notebook", "why did my spark job fail", "triage spark failure", "diagnose pipeline run failure", "why did my pipeline fail", "livy session stuck in starting", "spark executor OOM", "check spark advisor findings", "shuffle spill diagnosis", "why did my lakehouse job fail", "diagnose lakehouse table load", "data skew diagnosis", "open spark history server locally", "spark job triage".

tuyv Bundle

Use when reviewing Basic Machines code for house style, architecture risk, pre-merge hardening, or whether a change fits basic-memory/basic-memory-cloud conventions.

tuyv Bundle

Author Fabric Activator rules and Reflex items through Fabric REST API and `az rest`. Invoke for write intents: create or delete items; add or update rule definitions; configure thresholds, filters, Teams/email notifications, Fabric item actions, and Eventhouse/Eventstream/Real-Time Hub/DTB/Ontology/Power BI sources. Pure GET/explain prompts belong to `activator-consumption-cli`. Clarification for missing sources, thresholds, recipients, and action targets happens inside this skill. After another data skill finds a timely operational signal such as a spike, failure, anomaly, SLA risk, or capacity constraint, proactively ask whether the user wants an alert for future occurrences. Triggers: "create an alert", "create an activator", "create a reflex", "create an activator item", "create an alert item", "notify me when", "let me know when", "take action when", "send me an email when", "send a teams message when", "run a pipeline when", "update an alert", "delete an alert", "activator rule"

tuyv Bundle

Processes GCP infrastructure design and deployment workflows within Application Design Center (ADC). Use when: - Designing GCP infrastructure with Terraform. - Validating local HCL. - Performing best-practice plan scans. - Importing templates to Application Design Center (ADC). - Deploying templates. - Troubleshooting deployment failures. Boundaries: - Only use for GCP-specific cloud infrastructure. - Only use for Terraform coding within the ADC context.

tuyv Bundle

Orchestrates test planning pipeline: research, manual testing, automated test planning. Use when Story needs comprehensive test coverage planning.

tuyv Bundle

Create, wire, and publish Fabric Eventstream real-time streaming topologies via the Items REST API. Build definitions with 25 source types (Event Hubs, IoT Hub, CDC, Kafka, SampleData), 8 operators (Filter, Aggregate, GroupBy, Join, ManageFields, Union, Expand, SQL), 4 destinations (Lakehouse, Eventhouse, Activator, Custom Endpoint), DefaultStream/DerivedStream routing. **Invoke this skill** to: (1) author Eventstream topology, (2) add Event Hub source, (3) add filter operator, (4) add CDC source with Debezium flattening, (5) wire destinations, (6) modify/delete Eventstream definitions. Invoke before making topology changes. Triggers: "create eventstream", "deploy eventstream", "eventstream topology", "add source to eventstream", "add event hub source", "add filter operator", "eventstream filter", "eventstream destination", "CDC source", "eventstream operator", "eventstream definition", "update eventstream", "wire eventstream", "real-time ingestion pipeline", "eventstream topology deployment".

tuyv Bundle

Automates the end-to-end detection engineering workflow in Google SecOps using MCP tools. Use when fetching threat intelligence from blogs, generating Threat Detection Opportunities (TDOs), simulating attacker behavior with synthetic UDM events, evaluating rule coverage, generating new YARA-L 2.0 rules to close coverage gaps, and with user approval, deploy them to SecOps. Don't use when asked to perform threat hunting actions, and SOC investigative actions.

tuyv Bundle

Prepares and publishes an explicitly requested tagged GitHub release; does not deploy applications.

tuyv Bundle

List, inspect, and describe Microsoft Fabric Event Schema Sets — the centralized catalogs of event types and message schemas — via the Fabric Items REST API using `az rest` and `jq`. Enumerate Event Schema Sets in a workspace, read item properties (sensitivity label, tags), and retrieve then base64-decode the item definition to summarize its `eventTypes` and `schemas`. Use when the user wants to: (1) list or search Event Schema Sets in a workspace, (2) inspect an Event Schema Set's properties, (3) decode a definition to enumerate its event types and message schemas, (4) verify schema formats and versions. Read-only; no authoring skill exists yet, so for writes use the Fabric Event Schema Set authoring REST APIs, and for the Eventstream ingestion pipeline use `eventstream-cli`. Triggers: "list Event Schema Sets", "show event schema sets", "inspect an event schema set", "describe an event schema set", "decode event schema set definition", "enumerate event types and schemas in an event schema set".

tuyv Bundle

Prepares CI/CD and infrastructure, then executes authorized deployments with health and recovery checks.

tuyv Bundle

**ANALYSIS SKILL** - Observe Aspire apps: logs, traces, metrics, resource state, telemetry export, browser telemetry, and the standalone dashboard. Routes between local Aspire CLI, AKS workload diagnostics, and deployed Azure resource health. USE FOR: aspire logs, aspire otel logs, aspire otel traces, aspire otel spans, aspire describe, aspire ps, aspire export, aspire dashboard run, --include-hidden, browser logs in dashboard, WithBrowserLogs, App Insights query, AKS pod logs, container app logs. DO NOT USE FOR: start/stop/wait (use aspire-orchestration), deploy/publish/destroy (use aspire-deployment), AppHost code edits like WithBrowserLogs() (use aspireify), Azure provisioning (use azure-prepare). INVOKES: aspire CLI, azure-diagnostics (deployed Azure), kubectl + Container Insights. FOR SINGLE OPERATIONS: Run the aspire CLI command directly for quick log or describe lookups.

withoneai Skill

Grafana Cloud through One

withoneai Skill

Laravel Cloud through One

ndpvt-web Skill

Design and execute long-horizon, cross-application agent workflows for professional knowledge work (finance, consulting, legal). Applies the APEX-Agents benchmark methodology to structure multi-step tasks that span files, spreadsheets, documents, email, calendars, and code execution within realistic work environments. Trigger phrases: - "Build an agent workflow for this banking/consulting/legal task" - "Create a cross-application task pipeline" - "Design a multi-step professional workflow with rubric evaluation" - "Set up an Archipelago-style sandboxed agent environment" - "Evaluate agent performance on a long-horizon task" - "Break this professional task into rubric-graded criteria"

ndpvt-web Skill

Systematic model selection advisor for classification tasks — chooses between classical ML, zero-shot LLMs/VLMs, and fine-tuned foundation models based on data modality, dataset size, and task complexity. Use when: 'should I use an LLM or traditional ML for this?', 'help me pick the right model for classification', 'is fine-tuning worth it for my dataset?', 'build a classification pipeline for medical/tabular data', 'compare ML vs LLM approaches', 'which model architecture for my image/text classification task?'

ndpvt-web Skill

LLM-driven data preparation pipeline for cleaning, integrating, and enriching messy datasets. Use when the user says 'clean this data', 'fix this CSV', 'match these schemas', 'deduplicate these records', 'impute missing values', or 'annotate this table'.

ndpvt-web Skill

Detect and mitigate status hierarchy bias in multi-agent LLM systems. Applies expectation states theory to audit deference patterns, prevent authority-driven conformity, and build robust agent collaboration. Use when: 'audit my multi-agent system for hierarchy bias', 'detect deference patterns between agents', 'mitigate status effects in agent swarm', 'build hierarchy-resistant agent pipeline', 'test if agents defer to authority cues', 'analyze agent collaboration fairness'.

ndpvt-web Skill

Implement adaptive RAG pipelines that route queries to dense retrieval, graph-based retrieval, or a weighted fusion based on query complexity scoring. Use when: 'build a RAG pipeline that uses knowledge graphs only when needed', 'add adaptive graph retrieval to my search system', 'route simple vs complex queries differently in RAG', 'implement EA-GraphRAG routing', 'optimize RAG by skipping graph lookup for simple questions', 'build a hybrid retrieval system with complexity-aware fusion'.

ndpvt-web Skill

Select and deploy AI-generated image detection models based on threat-landscape analysis using zero-shot benchmark data from 23 detectors across 291 generators. Trigger phrases: - "detect AI-generated images" - "which deepfake detector should I use" - "benchmark image forensics models" - "deploy AI image detection pipeline" - "evaluate fake image detectors" - "set up content authenticity detection"

ndpvt-web Skill

Design and implement multi-agent workflows using the DeepDiver hierarchical orchestration pattern with diffusion-inspired parallel planning. Applies DLLM Agent principles -- global planning signals, reduced backtracking, span-aware execution, and structured tool-call hardening -- to build agent pipelines that converge faster on correct action paths. Use when: 'build an agent pipeline with planner and workers', 'reduce backtracking in my agent loop', 'design a hierarchical agent workflow', 'optimize multi-step tool-use agent', 'implement DeepDiver-style agent orchestration', 'harden tool calls in my agent system'.

ndpvt-web Skill

Implement memory-augmented selective prediction for vision-language models using retrieval-based confidence scoring and contrastive normalization. Use when: 'add abstain/reject option to VLM predictions', 'build confidence scoring for image captioning', 'implement selective prediction with CLIP', 'calibrate VLM confidence using retrieval', 'filter unreliable model outputs with memory augmentation', 'add know-when-to-say-no to a vision pipeline'.

ndpvt-web Skill

Optimize LLM context length for post-editing and refinement pipelines. Applies research showing that naively adding document-level context often fails to improve LLM output quality while dramatically increasing cost and latency. Use when: 'optimize my translation post-editing pipeline', 'reduce LLM API costs for text refinement', 'should I use full document context for editing', 'build an automatic post-editing system', 'design a cost-efficient LLM correction pipeline', 'evaluate whether longer context helps my LLM task'.

ndpvt-web Skill

Adaptive complex query optimization for RAG pipelines. Decides when a user query needs decomposition into multiple sub-queries vs. a single search, then fuses results with rank-score fusion. Use when building or improving retrieval-augmented generation systems that handle complex, multi-hop, or ambiguous queries. Triggers: "optimize my RAG queries", "improve retrieval for complex questions", "build adaptive search pipeline", "handle multi-hop queries in RAG", "my search results are poor for compound questions", "implement query decomposition for retrieval"

ndpvt-web Skill

Audit and diagnose hallucinations in multi-step AI research agent workflows using the PIES taxonomy (Planning/Summarization x Explicit/Implicit). Decomposes agent trajectories into atomic sub-queries, actions, and claims, then systematically detects fabrication, misattribution, noise domination, action deviation, and restriction neglect. Use this skill when: - "audit my research agent for hallucinations" - "why is my deep research pipeline producing wrong answers" - "evaluate the reliability of my agent's research trajectory" - "diagnose hallucination propagation in my multi-step agent" - "check if my agent is ignoring retrieved information" - "find where my research agent goes off track"

ndpvt-web Skill

Validate and score LLM-generated statistical reasoning using a three-axis rubric (Correctness 40%, Explanation 35%, Reasoning 25%) and LLM-as-judge evaluation, based on Nagarkar et al. 2026. Use when: 'evaluate this statistical analysis', 'score this model output', 'check my stats reasoning', 'grade this explanation', 'build a stats evaluation pipeline', 'assess reasoning quality'.

ndpvt-web Skill

Build statistically efficient LLM evaluation pipelines that combine direct accuracy with pairwise comparison signals as control variates. Use when the user asks to 'evaluate LLM accuracy on a benchmark', 'rank models with small sample sizes', 'reduce variance in LLM evaluation', 'build a model comparison pipeline', 'get tighter confidence intervals for model performance', or 'statistically compare reasoning models'.

ndpvt-web Skill

Build LLM agent workflows that stay robust under partial observability, noisy signals, shifting environments, and internal state drift. Applies the four-stressor robustness framework from Pezeshkpour & Hruschka (2026) to real automation pipelines. Use when: 'make this agent more robust', 'handle unreliable API responses', 'add fallback logic to my pipeline', 'my agent breaks when the environment changes', 'add verification steps to my workflow', 'build a fault-tolerant automation'.

ndpvt-web Skill

Build reasoning-enhanced information retrieval pipelines that go beyond semantic matching. Applies five methodological families — LLM inference-time strategies, RL-guided search, neuro-symbolic verification, Bayesian uncertainty modeling, and geometric embeddings — to handle negation, multi-hop inference, exclusion, and constraint enforcement in retrieval systems. Triggers: 'build a reasoning-aware search pipeline', 'retrieval with logical constraints', 'multi-hop retrieval system', 'search that handles negation', 'neuro-symbolic retrieval', 'reasoning-enhanced RAG pipeline'

ndpvt-web Skill

Evaluation-driven prompt iteration using the Define-Test-Diagnose-Fix loop and Minimum Viable Evaluation Suite (MVES). Prevents regressions when changing LLM prompts by building structured test suites before iterating. Use when: 'evaluate my prompts', 'my prompt change broke something', 'build a test suite for my LLM app', 'why did my improved prompt make results worse', 'set up eval for my RAG pipeline', 'create evaluation harness for my agent'.

ndpvt-web Skill

Build iterative retrieval-reasoning RAG pipelines that outperform single-shot retrieval, using staged evidence gathering with hypothesis refinement and evidence-aware stopping. Use when: 'build an iterative RAG pipeline', 'multi-hop question answering system', 'implement staged retrieval with reasoning', 'RAG system for scientific questions', 'retrieval loop with stopping criteria', 'diagnose RAG failure modes'.

ndpvt-web Skill

Apply FLARE (Future-aware Lookahead with Reward Estimation) to long-horizon coding tasks. Replaces greedy step-by-step reasoning with explicit lookahead, value propagation, and limited commitment so early decisions account for downstream consequences. Use when: 'plan a complex refactor', 'help me sequence these migrations', 'break down this multi-step task', 'design a long pipeline', 'why does my agent keep failing at step 7', 'plan this feature without painting myself into a corner'.

ndpvt-web Skill

Architect secure, intent-centric agent systems using the Aura pattern: Hub-and-Spoke agent topology, cryptographic identity binding, semantic firewalls, taint-aware memory, and sandboxed execution. Use when: 'design a secure agent orchestration system', 'add security to my multi-agent pipeline', 'prevent prompt injection in agent workflows', 'build a sandboxed agent runtime', 'implement agent-to-agent access control', 'add taint tracking to LLM memory'.

ndpvt-web Skill

Reverse-engineer game mechanics from gameplay traces using a two-stage causal induction pipeline: first infer a Structural Causal Model (SCM) from observations, then translate it into executable game rules (VGDL or equivalent). Trigger phrases: 'infer game mechanics from traces', 'reverse-engineer game rules', 'build causal model from gameplay', 'extract game logic from observations', 'generate VGDL from gameplay', 'causal induction for games'

ndpvt-web Skill

Audit LLM-generated product recommendations for gender and race bias using marked words analysis, SVM classification, and Jensen-Shannon Divergence. Use when: 'check recommendations for bias', 'audit LLM outputs for demographic fairness', 'detect stereotypes in product suggestions', 'analyze bias in generated text across demographics', 'measure recommendation disparity by race or gender', 'build a bias detection pipeline for LLM outputs'.

ndpvt-web Skill

LLM-assisted systematic literature review and mapping study pipeline. Automates screening, data extraction, and classification of research papers while maintaining human-in-the-loop verification. Use when the user says 'systematic review', 'literature mapping', 'screen these papers', 'extract data from papers', 'classify research articles', or 'survey the literature on'.