paulpas/agent-skill-router - Page 3

vercel-optimize

Optimizes Vercel-deployed applications for cost, performance, and reliability through cold start mitigation, caching strategy, edge function optimization, bundle auditing, and monitoring.

web-interface-guidelines

Applies comprehensive web interface design guidelines covering layout, typography, color, accessibility, responsive design, and UX patterns for consistent, user-friendly interfaces.

vercel-writing-guidelines

Enforces clear, concise, and user-focused writing guidelines covering tone, grammar, structure, formatting, and inclusive language for technical documentation and product copy.

agent-engineering-practices

Applies traditional software engineering rigor to AI agents through checkpoint/rollback state management, modular architecture, structured observability logging, and least-privilege permissions for production-grade reliability.

agent-maturity-levels

Provides a four-level maturity model (Core Reasoning → Connected Solver → Strategic Problem-Solver → Collaborative Multi-Agent) for assessing and planning agent architecture investments based on capability requirements.

deep-research-workflows

Orchestrates iterative autonomous research workflows combining multi-query exploration, gap analysis, follow-up refinement, and structured synthesis within configurable time budgets.

mass-framework

Automates multi-agent system design through three-stage optimization (block-level prompt tuning, influence-weighted topology search, workflow-level joint optimization) to discover optimal agent configurations and interactions.

agent-knowledge-graph

Constructs agent-centric knowledge graphs representing tools, capabilities, and their relationships for intelligent tool discovery during planning.

autonomous-coding-lifecycle

Manages the complete autonomous coding lifecycle from code generation through self-review, test validation, failure repair, and deployment with state-machine orchestration.

dual-provider-reasoning-routing

Routes reasoning workloads between multiple LLM providers based on task complexity, cost constraints, and latency requirements with real-time fallback chains.

framework-selection-matrix

Provides a decision matrix for selecting between agentic frameworks (LangChain, LangGraph, Google ADK, CrewAI) based on capability requirements, ecosystem fit, and production readiness.

iterative-prompt-refinement

Implements systematic prompt evolution through baseline generation, structured evaluation against rubrics, targeted revision, and regression validation to continuously improve prompt quality.

multi-platform-agent-deployment

Deploys and orchestrates an identical agent across multiple execution environments (CLI, web UI, API, chat platforms) with shared knowledge base and behavioral consistency.

structured-output-enforcement

Enforces deterministic agent output using JSON Schema, Pydantic models, and LLM-native format constraints to guarantee downstream consumers receive valid structured data.

verb-based-instruction-design

Decomposes agent instructions into action verbs with explicit success criteria per verb class to eliminate ambiguity and ensure LLMs correctly interpret intended operations.

budget-aware-cost-management

Implements session-level budget quotas, cost monitoring with threshold alerts, and ROI tracking to enforce AI agent spending limits and optimize return on investment.

cli-agent-workflows

Implements CLI agent workflows (terminal interaction, file operations, code generation from design specs, MCP bridging) for building command-line AI assistants and developer tooling.

coding-agent-frameworks

Implements autonomous coding agent frameworks (automated refactoring, test generation, deployment pipeline management) for AI-augmented software development with 30%+ code generation targets.

evaluation-rubrics-scored-review

Implements evaluation rubric design (multi-criteria scoring, Elo-based ranking, peer-review simulation) for quantitative assessment of AI agent outputs without ground truth labels.

knowledge-graph-construction

Implements knowledge graph construction (entity extraction, relationship mapping, graph database storage) and Graph-RAG integration for enterprise AI agents with structured reasoning over connected data.

reasoning-engine-internals

Implements dual-provider reasoning architecture (Gemini + Claude orchestration), cross-model reasoning pipelines, token budget management, and AI-assisted development velocity tracking for production AI systems.

a2a-communication

Implements Google's Agent-to-Agent (A2A) protocol for cross-framework agent communication using HTTP-based JSON-RPC 2.0 with agent card discovery, SSE streaming, and secure multi-agent interoperability patterns.

evaluation-monitoring

Continuously monitors agent performance in production through token tracking, LLM-as-a-Judge evaluation, A/B testing for improvements, drift detection, anomaly detection, and trajectory analysis with structured feedback loops.

exception-handling-recovery

Implements agent resilience patterns including retry logic with exponential backoff, fallback handler chains, state rollback, graceful degradation, and error escalation to maintain reliability under failure conditions.

exploration-discovery

Implements proactive agent exploration patterns for discovering unknown opportunities, generating hypotheses, designing experiments, and sharing findings through autonomous research loops beyond predefined optimization targets.

goal-setting-monitoring

Implements goal-oriented agent architectures with objective definition, LLM-based success criteria evaluation, iterative progress tracking, and max-iteration bounded refinement loops for proactive autonomous systems.

guardrails-safety

Protects agent systems from harmful outputs through behavioral constraints, input validation and sanitization, jailbreaking defenses, structured output enforcement, the Principle of Least Privilege, and fault-tolerant state management for safe autonomous…

gui-agent-interaction

Implements GUI agent interaction patterns (screen vision recognition, UI element detection, automated mouse/keyboard execution) for operating desktop and web applications without APIs.

human-in-the-loop

Integrates human oversight into AI agent workflows for high-stakes decisions through approval gates, feedback loops for RLHF, escalation policies, and decision augmentation patterns that balance automation with accountability.

learning-adaptation

Enables agents to improve through experience using reinforcement learning patterns (PPO, DPO, RLHF) and knowledge base RAG for continuous self-improvement and adaptive behavior across sessions.

mcp-integration

Integrates the Model Context Protocol (MCP) standard for LLM tool discovery and interaction, implementing MCP client-server architecture with stdio/HTTP transport, tool/resource/prompt types, and FastMCP SDK patterns.

memory-management

Manages agent memory across short-term (conversation buffers), long-term (vector stores, persistent databases), and procedural (learned patterns) layers to maintain stateful context across extended agent interactions.

multi-agent-collaboration

Orchestrates multiple specialized agents in concert using hierarchical, parallel, and sequential topologies (parent-child, debate/consensus, expert teams, sequential handoffs) to solve complex problems that exceed single-agent capability.

parallelization

Implements concurrent task execution patterns (parallel branches, fan-out/fan-in, multi-API calls, multi-modal processing) to reduce total agent processing time through independent subtask parallelism.

planning-patterns

Implements multi-step plan generation, iterative refinement, and dynamic task decomposition for proactive agent execution with self-correction.

prompt-chaining

Implements sequential prompt chaining patterns (linear LCEL pipelines, LangGraph stateful flows, Google ADK primitives) to decompose complex reasoning into reliable multi-step agent workflows.

prompt-engineering-patterns-v2

Implements advanced prompt engineering techniques (zero-shot/one-shot design, verb-based instructions, structured output, evaluation rubrics) for maximizing LLM response quality.

rag-patterns

Implements Retrieval-Augmented Generation patterns (chunking strategies, embedding-based vector search, semantic vs keyword retrieval, RAG pipelines) to ground LLM outputs in authoritative external knowledge sources.

reasoning-techniques

Implements advanced reasoning methodologies (Chain-of-Thought, Tree-of-Thoughts, ReAct, Self-Correction, Graph of Debates, Program-Aided LLMs) for multi-step problem-solving in complex agent tasks.

reflection-loop

Implements self-correction feedback loops (execution → evaluation/critique → refinement) to iteratively improve agent outputs through producer-critic collaboration and automated quality gates.