All authors
Claude Skills by Sathvikar01
github.com/Sathvikar0119 skills0 installs0 views
- Agent ArchitectureChoose the simplest AI-system architecture that correctly handles the task’s uncertainty. Use when designing or reviewing runtime autonomy, deciding whether a product needs an agent, or placing reasoning, tools, state and authority boundaries. Do not use for ordinary application architecture or executing a coding plan without an AI runtime design decision.Votes: 0GitHub stars: 2
- Agent Cost And LatencyOptimize AI-system cost and latency per verified successful outcome using measured budgets. Use when profiling model/tool/retrieval spend or delays, setting agent run ceilings, choosing model routing/escalation, caching, batching or parallelism under quality constraints. Do not use for general application performance without model/agent costs, speculative model recommendations, or reducing tokens without outcome measurements.Votes: 0GitHub stars: 2
- Agent Development WorkflowCoordinate the complete engineering lifecycle for substantial AI-agent or model-assisted workflow builds. Use when building, productionizing or substantially redesigning an agent, adding major tools/state/RAG/orchestration, or evaluating and hardening a whole agent system. Do not use for one bug, tiny prompt/schema/API changes, ordinary app work, simple deterministic utilities, conceptual agent explanations, or a narrow task owned by one specialist.Votes: 0GitHub stars: 2
- Agent EvalsDefine evaluation-first quality gates for an AI agent or model-powered workflow. Use when specifying agent success before implementation, building gold task sets, comparing agent variants, or diagnosing outcome and trajectory regressions. Do not use for ordinary unit-test writing or tool-interface usability alone without an agent quality measurement question.Votes: 0GitHub stars: 2
- Agent Failure RecoveryDesign bounded retries and durable recovery for AI-agent runs and side effects. Use when handling interrupted agent execution, ambiguous tool commits, partial completion, checkpoints/resume, poison tasks or retry/timeout policy. Do not use for general bug diagnosis, ordinary synchronous exception handling, or session handoff with no runtime recovery problem.Votes: 0GitHub stars: 2
- Agent GuardrailsAssemble and verify layered runtime guardrails for an AI system without confusing guidance with enforcement. Use when designing agent safety/quality containment, abstention gates, policy enforcement, permissions, runtime limits or approval layers, or reviewing a prompt-only guardrail claim. Do not use for security threat modeling alone, writing admission code alone, generic input validation, or making a system prompt sound safer.Votes: 0GitHub stars: 2
- Agent ObservabilityInstrument AI-agent trajectories so actions, evidence, state and quality can be reconstructed. Use when designing or reviewing traces/metrics for agent model calls, tools, approvals, retries, state, evals, cost or latency, or debugging an opaque agent run. Do not use for general service logging without an agent trajectory, full private reasoning collection, or quality scoring rules alone.Votes: 0GitHub stars: 2
- Agent OrchestrationDesign bounded next-step control for an AI runtime, from fixed routing to adaptive tool/retrieval choice and justified workers. Use when choosing who controls the next step, implementing agent routing, handoffs, fan-out/fan-in, shared state, cancellation or aggregate budgets. Do not use for coding-subagent dispatch, ordinary parallel utilities, whole-system lifecycle planning alone, or tool contracts without a runtime control-flow question.Votes: 0GitHub stars: 2
- Agent SecurityThreat-model and harden AI-agent trust boundaries, tools and data flows. Use when reviewing agent prompt injection, malicious retrieval/tool output, exfiltration, confused-deputy execution, tool poisoning or sandbox/credential boundaries. Do not use for general application security without a model/tool trust flow, or layering operational quality guardrails alone.Votes: 0GitHub stars: 2
- Agent State And MemoryDesign agent runtime state, checkpoints and memory without confusing recall with authoritative facts. Use when deciding what an AI system should persist, handling long-running context, cross-run memory, checkpoint/resume, provenance, expiry or deterministic state transitions. Do not use for coding-session context setup, a normal database migration, or conversational summarization with no runtime persistence design.Votes: 0GitHub stars: 2
- Agent TestingBuild a risk-based test pyramid and CI verification strategy for an AI-agent runtime. Use when implementing agent tests across deterministic admission/state, schemas, tool contracts, integration, trajectories, eval suites, adversarial behavior or restart/recovery. Do not use for ordinary test-first coding without an AI runtime, writing a gold-set scoring strategy alone, or browser automation unrelated to agent behavior.Votes: 0GitHub stars: 2
- Deterministic AuthorityBuild deterministic admission and authorization between model proposals and side effects. Use when an LLM can propose state changes, money movements, permission decisions or actions subject to executable constraints, or reviewing whether a model can override policy. Do not use for purely advisory text generation or ordinary business logic with no model-to-action trust boundary.Votes: 0GitHub stars: 2
- Evidence ProvenanceDesign and verify source-to-decision-to-output evidence lineage for AI systems. Use when implementing evidence ledgers, distinguishing retrieved from used support, reconciling decision/explanation/citation IDs, or preventing fabricated, stale or spoofed provenance across tools and model stages. Do not use for retrieval ranking/chunking alone, ordinary bibliography formatting, general logging, coding-source lookup, or answers with no evidence-lineage contract.Votes: 0GitHub stars: 2
- Human In The LoopPlace proportional human review and approval gates in an AI system's execution path. Use when designing approval/escalation for agent actions with financial, external, destructive or permission impact, preview-before-commit flows, or uncertainty that warrants human judgment. Do not use for requesting routine coding confirmation, imposing approval on harmless reads, or deterministic authorization logic alone.Votes: 0GitHub stars: 2
- Prompt EngineeringDesign concise, versioned production prompts with responsibilities, trust boundaries and measurable behavior. Use when building or revising deployed agent/workflow prompts, defining instruction hierarchy, tool/output guidance, uncertainty or termination, or evaluating a prompt regression. Do not use for generic prompt tricks, one-off prose requests, repository context setup, or using a prompt as authorization/security enforcement.Votes: 0GitHub stars: 2
- Rag EngineeringDesign, diagnose and evaluate retrieval-grounded generation with evidence provenance. Use when building or improving document-backed AI answers, measuring retrieval quality, investigating unsupported citations, or deciding whether chunking, hybrid search, reranking or query transformation helps. Do not use for ordinary database search without generation, general browsing, or memory persistence design without a retrieval-quality problem.Votes: 0GitHub stars: 2
- Structured Output DesignDesign and validate model-produced structured proposals before application use. Use when specifying JSON or typed outputs from an LLM, handling refusals or malformed generations, validating semantic constraints, or migrating model output contracts. Do not use for routine API serialization or database schema design without model-generated data.Votes: 0GitHub stars: 2
- Tool DesignDesign model-facing tools with coherent capabilities, strict contracts and predictable side effects. Use when exposing APIs or operations to an LLM agent, redesigning ambiguous agent tools, or specifying tool names, descriptions, arguments, results and execution semantics. Do not use for ordinary REST design with no model consumer, MCP transport implementation alone, or evaluating tool selection without changing contracts.Votes: 0GitHub stars: 2
- Tool EvalsMeasure whether an LLM agent can discover, select, compose and recover with a tool surface. Use when testing model-facing tool usability, argument accuracy, response interpretation, multi-step tool chains, unnecessary calls or tool failure handling. Do not use for backend API contract tests alone, MCP server scaffolding, or whole-agent answer-quality benchmarking without a tool-use question.Votes: 0GitHub stars: 2