Claude Skills by viktorbezdek

This skill should be used when the user asks to "evaluate agent performance", "build test framework", "measure agent quality", "create evaluation rubrics", "implement LLM-as-judge", "compare model outputs", "mitigate evaluation bias", or mentions multi-dimensional evaluation, agent testing, quality gates, direct scoring, pairwise comparison, position bias, evaluation pipelines, or automated quality assessment for LLM agent systems. NOT for testing code or applications (use testing-framework),...

This skill should be used when the user asks to "start an LLM project", "design batch pipeline", "evaluate task-model fit", "structure agent project", or mentions pipeline architecture, agent-assisted development, cost estimation, or choosing between LLM and traditional approaches. NOT for evaluating agent quality or building evaluation rubrics (use agent-evaluation), NOT for multi-agent coordination or agent handoffs (use multi-agent-patterns).

Design production-grade REST, GraphQL, gRPC, and Python library APIs with correct schemas, error contracts, auth, and versioning. Use when the user asks to design an API, define endpoints, create an OpenAPI/Swagger spec, design a GraphQL schema, build a gRPC service, model request/response with Pydantic, add pagination, or review API contracts. NOT for building MCP server tools (use mcp-server). NOT for Node.js/Express API routes or backend patterns (use backend-patterns or typescript-develop...

This skill should be used when the user asks to "model agent mental states", "implement BDI architecture", "create belief-desire-intention models", "transform RDF to beliefs", "build cognitive agent", or mentions BDI ontology, mental state modeling, rational agency, or neuro-symbolic AI integration. NOT for multi-agent coordination or agent handoffs (use multi-agent-patterns), NOT for agent memory frameworks or persistence (use memory-systems).

Design ad-hoc personas for niche domains when the canonical 12 brainstorm- swarm personas don't fit. Covers: when a custom persona is justified (vs forcing canonical to fit), the persona-design template (voice, contribution shape, output format), anti-patterns (too-narrow personas, redundant personas, sock-puppet personas), and how to invoke a custom persona inline via Task() with a tailored prompt rather than a saved subagent definition. Use when running a brainstorm-swarm and the topic call...

Structure the interview arc when a brainstorm-swarm interviews the user. Covers the divergent-then-convergent arc (open with breadth, close with depth), question design (open vs probing, leading vs neutral, what-vs-why- vs-how), depth-vs-breadth tradeoffs, and when to send a second round of follow-up questions to specific personas. Use when running a brainstorm swarm and the swarm-protocol skill has spawned the persona subagents — this skill teaches how those personas interact with the user d...

Orchestration logic for running a parallel persona-swarm brainstorm — when to invoke, which subset of the 12 canonical personas to spawn (PM, Engineer, Designer, Skeptic, User Advocate, Pre-Mortem Specialist, Junior, Veteran, First-Principles Thinker, Constraint-Setter, Optimist, Operator), how to spawn them in parallel via the Task() tool with persona-specific subagent types, how to handle their outputs, and when to do a second round. Use when the user asks to brainstorm with multiple perspe...

Combine a parallel persona-swarm's outputs into an actionable artifact — consensus matrix (what every persona agreed on), dissent log (where personas disagreed and why), open questions (what nobody could answer), recommended next move (synthesized decision). Preserves dissent rather than forcing consensus. Use when running a brainstorm-swarm and the swarm-protocol skill has collected the persona outputs — this skill produces the synthesis artifact. NOT for orchestrating the spawn (use swarm-p...

CI/CD pipeline design and DevOps automation — use when the user mentions GitHub Actions, GitLab CI, Jenkins, Terraform, infrastructure as code, DevSecOps, ArgoCD, Kubernetes deployment automation, or pipeline configuration YAML. NOT for release orchestration or semantic-release workflows (use git-workflow), NOT for Docker containers or Dockerfiles (use docker-containerization), NOT for git branching or commits (use git-workflow).

Expert FinOps guidance covering cloud, AI, SaaS, and adjacent technology spend. Includes AI cost management, GenAI capacity planning, AI-powered FinOps automation, Anthropic billing, AWS (EC2, Bedrock, Savings Plans, CUR, commitment strategy), Azure (reservations, Savings Plans, AHB, OpenAI PTUs, portfolio liquidity), GCP (Vertex AI, Compute Engine, BigQuery), Kubernetes and container FinOps (OpenCost, Kubecost), serverless FinOps (Lambda, Functions, Cloud Run), data platforms (Kafka/MSK, Ela...

Reviews existing code and pull requests using multi-agent swarm analysis covering security, performance, style, test coverage, and documentation quality. Extracts and prioritizes PR comments, performs security audits, and generates actionable fix plans with file:line references. Use when the user asks to review code, review a PR, audit code for security, assess code quality, analyze pull request comments, get feedback on existing code, or perform a code audit. NOT for writing new code or impl...

Edit written text for clarity and conciseness — active voice, hedge and weasel-word removal, jargon strip, sentence compression, nominalization fixes, and readability. Use when the user asks to tighten, shorten, edit, or clarify a paragraph or doc, remove hedges and weasel words, convert to active voice, cut jargon, kill passive voice, fix nominalizations, or improve readability. NOT for structuring the overall doc (use structured-writing). NOT for UI microcopy (use ux-writing). NOT for auto-...

Decide what and when to write down, and produce Architecture Decision Records (ADRs), one-pagers, runbooks, and decision logs. Use when the user asks whether something should be documented, wants to write an ADR, needs a runbook or operational playbook, wants to keep a decision log, or is trying to decide between a one-pager and an RFC. NOT for auto-generating docs from code (use documentation-generator). NOT for end-user tutorials or how-tos (use example-design). NOT for docs-site informatio...

Align stakeholders in writing using RFCs, design docs, proposals, pre-reads, and decision docs with explicit role assignments (DACI, RAPID). Use when the user asks to write an RFC, design doc, proposal, pre-read, or decision doc, wants to align async stakeholders on a decision, needs to assign deciders vs consulted vs informed, wants to structure a cross-team proposal, or is preparing a pre-read for a decision meeting. NOT for structuring a generic memo (use structured-writing). NOT for stake...

Structure a written piece using BLUF (Bottom Line Up Front), the Minto Pyramid Principle, inverted pyramid, or SPQR (Situation-Problem-Question- Resolution). Use when the user asks to write BLUF-style, apply the Pyramid Principle, structure a memo/email/doc, lead with the conclusion, outline a long message so it is skim-readable, or reorganize buried-lede writing. NOT for line-level clarity and concision editing (use clarity-editing). NOT for UI microcopy (use ux-writing). NOT for narrative o...

Communicate systems and flows visually using Mermaid (flowchart, sequence, state, ER, class), the C4 model, and diagram-as-code. Use when the user asks to draw a flow, sequence diagram, state machine, ER diagram, C4 model, architecture diagram, or wants Mermaid code for a system, wants to pick a diagram type, or wants to visualize a process as a flowchart. NOT for UI/UX mockups or visual design (use frontend-design). NOT for slide graphics or presentations (use frontend-slides). NOT for wiref...

Establish and enforce uniform naming conventions, taxonomy standards, style guides, and content reuse patterns across a project. Use when the user asks to audit for consistency, standardize naming, create a style guide, align terminology across docs, eliminate drift, or define reuse patterns across content or code. NOT for formal knowledge graphs or semantic ontologies (use ontology-design). NOT for CMS content types or editorial workflows (use content-modelling). NOT for language-specific co...

Design CMS content models — content types, fields, editorial workflows, governance rules, and COPE (Create Once, Publish Everywhere) patterns — for structured, multi-channel publishing. Use when the user asks to design a content model, define content types in a CMS, structure fields for editorial content, plan a headless CMS architecture, or design content reuse across channels. NOT for formal knowledge graphs, OWL/RDF ontologies, or semantic modeling (use ontology-design). NOT for naming con...

REDUCING context size — summarization strategies, anchored iterative summarization, tokens-per-task optimization, compaction triggers, and probe-based evaluation. Use when the user asks to "compress context", "summarize conversation history", "implement compaction", "reduce token usage", or mentions structured summarization or long-running sessions exceeding context limits. NOT for diagnosing context failures or degradation patterns (use context-degradation), NOT for KV-cache optimization or ...

Diagnosing context FAILURES — lost-in-middle, poisoning, distraction, confusion, and clash patterns with model-agnostic measurement workflows. Use when the user asks to "diagnose context problems", "fix lost-in-middle issues", "debug agent failures", "understand context poisoning", or mentions context degradation, context clash, or agent performance degradation. NOT for learning context basics or theory (use context-fundamentals), NOT for compressing or summarizing context (use context-compre...

Foundational theory of context engineering — what context IS, how attention works, progressive disclosure principles, and context budgeting basics. Use when the user asks to "understand context", "explain context windows", "learn context engineering", or discusses context components, attention mechanics, or context budgets. NOT for fixing broken context or diagnosing failures (use context-degradation), NOT for compressing or summarizing context (use context-compression), NOT for KV-cache or p...

EXTENDING effective context capacity — KV-cache optimization, observation masking, context partitioning, and retrieval strategies. Use when the user asks to "optimize context", "implement KV-cache", "partition context", "mask observations", or mentions extending context capacity or cache-friendly prompt design. NOT for reducing or compressing content via summarization (use context-compression), NOT for diagnosing context failures or degradation patterns (use context-degradation), NOT for file...

Generate new ideas and breakthrough solutions using brainstorming, lateral thinking, SCAMPER, first principles reasoning, game theory, and strategic reframing. Use when the user asks to brainstorm, ideate, "think outside the box", generate alternatives, explore creative solutions, reframe a problem, or come up with ideas — especially when stuck in conventional thinking. NOT for stress-testing or finding flaws in existing ideas (use critical-intuition), NOT for making a final decision between ...

Stress-test, critique, and challenge existing ideas through pattern recognition, bias detection, Bayesian reasoning, blind-spot exposure, and red-flag identification. Use when the user asks to stress-test a plan, challenge assumptions, find flaws, identify risks in an idea, expose hidden biases, audit a proposal for blind spots, or "poke holes in this". NOT for generating new ideas or solutions (use creative-problem-solving), NOT for structured risk registers or mitigation plans (use risk-man...

Finds and fixes bugs through systematic root cause analysis, stack trace interpretation, browser DevTools automation, CI/CD pipeline debugging, performance profiling, test pollution detection, and AI-powered error analysis. Use when the user asks to debug, fix a bug, investigate an error, analyze a stack trace, find root cause of a failure, profile performance, diagnose test failures (unit/integration/E2E), troubleshoot CI/CD pipelines, debug flaky tests, use Chrome DevTools, or trace data fl...

Docker and container development — use when the user mentions Dockerfiles, multi-stage builds, Docker Compose, container optimization, image size reduction, DDEV, containerization, or dev environment setup with containers. NOT for CI/CD pipeline YAML or pipeline configuration (use cicd-pipelines), NOT for workflow orchestration or release automation (use workflow-automation), NOT for Kubernetes or container orchestration platforms (use cloud-native tooling).

Generate comprehensive documentation for a codebase by reading the repository and producing READMEs, API docs, architecture docs, and technical references. Use when the user asks to "document this repo", "generate docs", "write a README", "create API documentation", "document this codebase", "write architecture docs", or "produce technical references" for an existing project. NOT for UX copy, button labels, or interface microcopy (use ux-writing). NOT for pedagogical code examples or tutorial...

Identify and document boundary conditions, corner cases, error scenarios, and validation requirements that implementations must handle. Use when the user asks to find edge cases, identify corner cases, specify validation rules, enumerate error scenarios, harden a function against bad inputs, or think through what can go wrong at the boundaries of a system. NOT for writing the actual tests (use testing-framework or test-driven-development). NOT for structured risk registers around project-leve...

Psychological elicitation and deep-interview design using narrative identity (McAdams), self-defining memories (Singer), Motivational Interviewing (Miller & Rollnick OARS), values elicitation (Schwartz), schema detection (Young), and life review (Haight/Birren). Use when designing user interviews that need to reveal motivations rather than stated preferences, writing conversation flows for personal discovery, auditing interview scripts for interrogation anti-patterns, building conversational ...

Design pedagogically effective code examples, tutorials, and runnable samples using progressive complexity and deliberate scaffolding. Use when the user asks to write a code example that teaches a concept, design a quickstart tutorial, create sample code for a library or API, build a runnable demo, or structure examples from simple to advanced. NOT for generating full repo documentation or READMEs (use documentation-generator). NOT for writing examples inside a skill file (use skill-foundry)....

Using the FILE SYSTEM for context — scratch pads, plan persistence, dynamic skill loading, sub-agent file workspaces, and terminal log persistence. Use when the user asks to "offload context to files", "implement scratch pads", "persist agent plans", "use filesystem for agent memory", or mentions file-based context management, tool output persistence, or just-in-time context loading. NOT for in-context optimization like KV-cache or observation masking (use context-optimization), NOT for summa...

Visual design systems, UI/UX styling, Tailwind CSS, CSS variables, component libraries (shadcn/ui, Radix UI), design tokens, accessibility (WCAG), responsive layout, dark mode, and Figma-to-code workflows. NOT for React component logic, hooks, or state management (use react-development). NOT for Next.js routing, SSR, or server components (use nextjs-development).

Git workflow management — use when the user mentions git, conventional commits, commit quality, branch management, worktree operations, GitFlow, changelog generation, semantic versioning, release notes, backlog management, or issue tracking integration. NOT for CI/CD pipelines or pipeline YAML (use cicd-pipelines), NOT for non-git workflow orchestration (use skillstack-workflows or multi-agent-patterns), NOT for code review content or PR quality assessment (use code-review).

Google Workspace CLI (gws) for managing all 18 Workspace APIs from the terminal. Use when running gws commands, listing Drive files, sending Gmail, reading Sheets, creating Calendar events, managing Tasks, querying Chat, pushing Apps Script, building cross-service automations, or when user mentions gws or Google Workspace CLI. NOT for building MCP servers (use mcp-server), NOT for general API design (use api-design), NOT for GCP infrastructure (use cicd-pipelines).

Build and deploy hosted background coding agents with sandboxed VM execution, multiplayer sessions, and multi-client interfaces. Use when the user asks to "build background agent", "create hosted coding agent", "set up sandboxed execution", "implement multiplayer agent", or mentions background agents, sandboxed VMs, agent infrastructure, Modal sandboxes, self-spawning agents, or remote coding environments. NOT for agent coordination patterns or multi-agent design (use multi-agent-patterns), N...

MCP (Model Context Protocol) server development — use when the user mentions MCP, Model Context Protocol, FastMCP, MCP server, MCP tool, Claude Code plugin, or building agent tools with MCP. Covers server implementation in Python or TypeScript, evaluation testing, production deployment, and plugin packaging. NOT for designing tool interfaces or tool consolidation patterns for agents (use tool-design), NOT for prompt engineering or prompt optimization (use prompt-engineering).

Template for creating Claude Code skills. Replace this description with a clear explanation of what this skill does and when it should be used.

Guides implementation of agent memory systems, compares production frameworks (Mem0, Zep/Graphiti, Letta, LangMem, Cognee), and designs persistence architectures for cross-session knowledge retention. Use when the user asks to "implement agent memory", "persist state across sessions", "build knowledge graph for agents", "track entities over time", "add long-term memory", "choose a memory framework", or mentions temporal knowledge graphs, vector stores, entity memory, adaptive memory, dynamic ...

This skill should be used when the user asks to "design multi-agent system", "implement supervisor pattern", "create swarm architecture", "coordinate multiple agents", or mentions multi-agent patterns, context isolation, agent handoffs, sub-agents, or parallel agent execution. NOT for agent memory or persistence (use memory-systems), NOT for tool design or tool interfaces (use tool-design), NOT for hosted agent infrastructure or sandboxed VMs (use hosted-agents), NOT for BDI cognitive models ...

Design information architecture, wayfinding systems, and navigation structures for documentation sites and applications. Use when the user asks to design navigation, plan a sitemap, structure a sidebar or menu, define content hierarchy, design breadcrumbs, or organize how users move through content. NOT for user journey maps or touchpoint flows across time (use user-journey-design). NOT for microcopy, labels, or button text in navigation (use ux-writing). NOT for Next.js routing, layouts, or ...

Next.js framework development including App Router, Server Components, Server Actions, SSR, SSG, ISR, caching, data fetching, middleware, layouts, parallel routes, and module architecture for Next.js 13+/15/16. NOT for generic React patterns, hooks, or component logic (use react-development). NOT for UI/CSS design systems or visual styling (use frontend-design).

Design formal knowledge models — classes, properties, relationships, hierarchies, and semantic graphs — for knowledge representation and reasoning. Use when the user asks to build an ontology, design a knowledge graph, model entity relationships formally, define class hierarchies, create a taxonomy for semantic reasoning, or structure data for RDF/OWL. NOT for CMS content types, editorial workflows, or publishing structures (use content-modelling). NOT for naming conventions or terminology st...

Conduct deep OSINT research on individuals — from name or handle to a scored dossier with psychoprofile (MBTI/Big Five), career map, and confidence-graded facts. Phased pipeline (0→6): tooling check, seed collection, internal intelligence, platform extraction, cross-reference, psychoprofile, completeness evaluation, dossier output. Swarm mode: 3-5 parallel Sonnet sub-agents. 55+ Apify actors. 7 search APIs. Trigger phrases: "osint", "research person", "find everything about", "due diligence",...

Reframe work around measurable outcomes using OKRs, KPIs, and the outcome-vs-output distinction. Use when the user asks to define success criteria, write OKRs, set KPIs, clarify what "done" means in terms of impact, distinguish outputs from outcomes, or make a goal statement measurable and time-bound. NOT for ranking or scoring features by priority (use prioritization). NOT for systemic feedback-loop analysis (use systems-thinking).

Create individual user personas and customer archetypes — with demographics, goals, pain points, behaviors, and empathy maps — to represent the humans a product or system is built for. Use when the user asks to define personas, create user archetypes, describe target users, build empathy maps, or characterize the audience for a product or design decision. NOT for mapping stakeholders across an organization, RACI charts, or influence analysis (use persona-mapping). NOT for designing research i...

Map stakeholders across organizations using Power-Interest matrices, RACI charts, influence analysis, and salience models to understand who has authority, who is affected, and how to engage each group. Use when the user asks to map stakeholders, create a RACI matrix, analyze organizational influence, identify decision-makers, prioritize who to involve, or plan stakeholder communication. NOT for individual user personas or empathy maps for product design (use persona-definition). NOT for desig...

Decides which Claude Code extension type to use for a given capability — skill, hook, MCP server, subagent, or slash command — and designs the plugin.json manifest around that decision. Use when designing a plugin, planning plugin structure, choosing between skill vs hook vs mcp, setting up plugin composition, laying out plugin components, or refactoring an existing plugin whose decomposition feels wrong. Covers the five extension types, the decision matrix (when each fits), plugin.json schem...

Integrates multiple components inside a single Claude Code plugin — skills, hooks, MCPs, subagents, and commands — using the canonical directory layout and path substitution variables. Use when building a multi-component plugin, combining skills and hooks, laying out the plugin directory structure, working with CLAUDE_PLUGIN_ROOT or CLAUDE_PLUGIN_DATA, choosing where to put shared scripts inside a plugin, or debugging why a path reference is not resolving at runtime. Covers the canonical dire...

Generates comprehensive documentation for any Claude Code plugin by fetching and analyzing all source files. Use when you need to document a Claude plugin, write a plugin README, explain how a plugin works, generate plugin usage guides, create plugin tutorials, analyze a plugin architecture for documentation, or produce installation and setup instructions. Takes a GitHub URL or local path, fetches all plugin files, maps the component architecture, identifies real use-case scenarios from skill...

Measures whether a Claude Code plugin actually works by running triggering evals (does the model pick the skill?) and output evals (does it produce correct results?). Use when you need to evaluate a plugin, run skill activation testing, set up an eval harness, measure plugin quality, write trigger rate tests, check output quality, compare plugin iterations, or iterate on a SKILL description based on eval results. Covers the two eval file formats (trigger-evals.json and evals.json), the grader...