Marketplace

Invoke when an autonomous task may outlast one context window—persist resumable checkpoints to a state file so work survives resets and interruptions.

Use when planning a provider-to-provider cloud migration: service equivalence map, egress cost model, cutover runbook, rollback plan and lock-in scoring.

Design and configure monorepo structures: workspace setup, package boundaries, dependency management, and CI optimization.

Build comprehensive model evaluation: benchmark selection, statistical significance, human evaluation protocols, and safety testing.

Plan ML model deployment: serving patterns, A/B testing, canary releases, monitoring, and rollback for production ML.

Plan and execute safe database schema migrations: dependency ordering, rollback strategies, zero-downtime patterns, and data backfill procedures.

Use when asked to build or review an MCP server wrapping an API, database, or CLI so agents can call it through well-designed, strictly typed tools.

Design Model Context Protocol resources and prompts that expose data and workflows to AI agents following the MCP specification.

Plan and execute LLM fine-tuning: dataset preparation, method selection (LoRA, QLoRA, full), evaluation, and deployment.

Use when building or auditing an LLM-as-judge: calibrate it against human labels, measure agreement, and de-bias position and verbosity effects.

Systematically modernize legacy codebases: strangler fig pattern, incremental rewrites, and risk-minimized migration paths.

Invoke to audit or tighten an AI agent's tool/MCP permissions: diff granted access against observed usage and recommend a least-privilege allowlist.

Build knowledge graphs from unstructured data for agent grounding: entity extraction, relation mapping, and graph-based retrieval.

Use when reviewing or writing Kubernetes manifests; audits workloads for reliability, security, and cost issues and produces ready-to-apply fixes.

Optimize model inference: quantization, distillation, batching, speculative decoding, and serving infrastructure tuning.

Design incident response playbooks: detection triggers, severity classification, containment, eradication, and communication.

Use after an incident is resolved to produce a blameless postmortem with a reconstructed timeline, contributing factors, and owned action items.

Design least-privilege IAM policies for AWS, GCP, and Azure: role design, permission boundaries, and access auditing.

Design reusable Infrastructure as Code modules: Terraform, Pulumi, and CloudFormation patterns for maintainable infrastructure.

Use when provisioning, hardening or cost-modelling Hetzner cloud and dedicated servers: hcloud, Robot API, private networks, floating IPs, backups.

Drive headless browsers (Playwright/Puppeteer) to navigate pages, auto-screenshot at each step, fill forms, click elements, and capture visual evidence of workflows.

Use when auditing an LLM answer that carries source citations, to verify each claim is entailed by its cited sources and flag unsupported or misattributed ones.

Design and implement safety guardrails for AI agents: input/output filtering, topic restrictions, PII redaction, and jailbreak resistance.

Design GraphQL schemas: type design, N+1 prevention, pagination patterns, mutation design, and schema evolution.