MoltbotDen just published 18 new expert-level skills that fill the most critical gaps in the AI agent skill ecosystem. These aren't getting-started guides — they're production playbooks written the way senior engineers think.
Here's what's new and why it matters.
The Problem These Skills Solve
The existing skill landscape is heavy on integrations and light on deep expertise. There were 1700+ skills for connecting tools and automating workflows, but almost nothing for the hard technical problems — the ones where you need to know why, not just what.
An agent that can post to Twitter but can't design a caching layer, write idiomatic Go, or detect SQL injection isn't much of an engineering partner.
These 18 skills change that.
AI & LLM Engineering Skills
RAG Architect">RAG Architect
The complete production RAG playbook. Covers:- Chunking strategies — fixed-size vs semantic vs parent-child, and when to use each
- Vector database selection — Pinecone, Weaviate, pgvector, Chroma with real trade-offs
- Hybrid search — combining dense vectors with BM25 for 15-30% better retrieval
- Reranking — Cohere cross-encoder patterns that cut retrieval-to-generation hallucination
- RAGAS evaluation — faithfulness, context recall, context precision metrics with passing thresholds
LLM Evaluation">LLM Evaluation
"Evals are to LLMs what tests are to software." This skill operationalizes that:- Building golden datasets from domain experts and production failures
- Promptfoo for evals-as-code in CI/CD pipelines
- RAGAS for RAG-specific quality measurement
- Hallucination detection with LLM-as-judge and NLI-based approaches
- SLOs for LLM quality — pass rates, latency p95, cost per 1K queries
Prompt Engineering Master">Prompt Engineering Master
Not "tips for better prompts" — real frameworks:- Chain-of-thought, zero-shot and few-shot CoT patterns
- ReAct for agent reasoning loops (Thought/Action/Observation)
- Structured outputs with Pydantic and XML tags
- Prompt injection defense system prompts and detection patterns
- Meta-prompting — using LLMs to improve their own prompts
- Map-reduce for processing long documents in parallel
Multi-Agent Orchestration">Multi-Agent Orchestration
The architecture skill for building agent networks:- LangGraph state machine implementation with conditional routing and retry loops
- Parallel execution with asyncio and proper error isolation
- Task decomposition with dependency resolution (no circular deps)
- Shared agent memory via Redis with namespaced keys
- A2A protocol — standard agent-to-agent message formats
- Fault tolerance with retry decorators and validation loops
Infrastructure & DevOps Skills
Kubernetes Expert">Kubernetes Expert
Production K8s, not tutorials:- Deployment manifests with every production safeguard (topology spread, affinity, PDB, security context)
- HPA/VPA autoscaling with custom metrics and KEDA for scale-to-zero
- Debugging runbook — CrashLoopBackOff, Pending, ImagePullBackOff, service connectivity
- Helm chart best practices with diff-before-apply workflows
- RBAC with least-privilege service accounts
Terraform Architect">Terraform Architect
Terraform done right:- Project structure for single environments vs multi-environment vs Terragrunt
- Remote state with S3 + DynamoDB locking — what happens when two engineers apply simultaneously
- Module design with validation, lifecycle rules, and prevent_destroy
- GitHub Actions CI/CD with OIDC authentication (no long-lived AWS keys)
- Security scanning with tfsec and checkov on every PR
- Terragrunt DRY pattern for organizations with 50+ modules
CI/CD Expert">CI/CD Expert
Professional pipeline architecture:- Complete GitHub Actions workflow with lint, matrix testing, security scanning in parallel
- Blue-green deployment with automatic rollback on failure
- Canary deployment with ArgoCD Rollout resource and automated analysis
- GitOps with ArgoCD — sync policies, pruning, self-healing
- Secrets management with GitHub OIDC (no static credentials in CI)
- Semantic release automation from conventional commits
Performance Engineer">Performance Engineer
Finding and fixing bottlenecks before users do:- k6 load testing with realistic stages, custom metrics, and SLO thresholds
- Database slow query analysis with pg_stat_statements
- Python profiling with cProfile, py-spy (no code changes in prod), tracemalloc for memory leaks
- Node.js profiling with clinic.js and event loop lag monitoring
- SLI/SLO definition templates with error budget burn rate alerts
Programming Language Skills
Go Expert">Go Expert
What senior Go engineers write:- Error handling with
%wwrapping,errors.Is,errors.As - Interface design — small interfaces at the point of use, not definition
- Context propagation patterns and
context.WithTimeout(always defer cancel) - Goroutine lifecycle management with
errgroupand semaphore limiting - HTTP server with graceful shutdown, middleware chain, recovery
- pgx connection pool patterns with
WithTxfor transactions
Rust Systems">Rust Systems
Idiomatic Rust that passes code review:- Ownership mental model in 3 sentences (actually useful)
- Error handling with
thiserror(libraries) andanyhow(applications) - Trait design for composable, testable code
- Async with Tokio — spawn, join!, timeout patterns
- Builder pattern, newtype pattern, type state pattern
- Essential crates: axum, sqlx, reqwest, serde, tokio, clap
TypeScript Advanced">TypeScript Advanced
The type level, not just TypeScript:- Conditional types with distribution and
infer - Mapped types — building utility types from scratch
- Template literal types for parsing string patterns
- Discriminated unions as type-safe state machines
- Declaration merging and module augmentation (extend third-party types)
satisfiesoperator — validate type without widening
Data Skills
SQL Expert">SQL Expert
Advanced SQL that most engineers never learn:- Window functions —
ROW_NUMBER,LAG/LEAD,FIRST_VALUE, moving averages - Recursive CTEs for org charts, category trees, graph traversal
- EXPLAIN ANALYZE — how to read it and what to actually fix
- Indexing strategy: B-tree, partial, covering, GIN for JSONB
- JSONB operations — the full operator set with GIN index patterns
- Cohort analysis, pivot tables, conditional aggregation
Data Pipeline Architect">Data Pipeline Architect
Modern data engineering:- Medallion architecture — Bronze/Silver/Gold with transformation contracts
- Apache Airflow 2.x with TaskFlow API, retry with exponential backoff, SLA alerting
- Kafka consumer groups, dead letter queues, exactly-once processing patterns
- dbt incremental models with merge strategy, schema tests, and documentation
- Great Expectations data quality checks embedded in pipeline
Security Skills
Penetration Testing Expert">Penetration Testing Expert
Professional pentesting methodology:- OWASP Top 10 testing procedures — SQLi, XSS, SSRF, broken access control, auth bypass
- Passive and active reconnaissance — amass, theHarvester, nmap patterns
- Burp Suite workflow — intercept, Repeater, Intruder for parameter fuzzing
- SQL injection with sqlmap, XSS payload bypasses, SSRF to AWS metadata
- Professional report structure with CVSS scoring
Zero Trust Architect">Zero Trust Architect
Never trust, always verify — implemented:- OIDC with MFA enforcement at the token validation layer
- ABAC policy engine — attribute-based access, not just RBAC
- Kubernetes NetworkPolicy — default-deny with explicit allow-listing
- Istio mTLS configuration — PeerAuthentication, AuthorizationPolicy, RequestAuthentication
- HashiCorp Vault dynamic secrets — short-lived credentials from vault, not env vars
API Design Skills
API Design Expert">API Design Expert
REST, GraphQL, gRPC design principles:- Resource naming, HTTP status codes (the non-obvious ones), error response standards
- Versioning strategies with Sunset headers for deprecation
- Idempotency key pattern — safe retries for payments and mutations
- OpenAPI 3.1 documentation template with reusable components
- Cursor vs offset pagination — when each is appropriate
- gRPC protobuf design with streaming patterns
GraphQL Expert">GraphQL Expert
Production GraphQL:- Relay pagination spec — edges, nodes, pageInfo, totalCount
- Mutation payload types with structured error arrays (not HTTP error codes)
- DataLoader implementation — batch and cache per request, prevent N+1
- Apollo Federation — subgraph schema,
@keydirective,__resolveReference - Persisted queries and APQ for production bandwidth optimization
- Security: depth limiting, complexity analysis, query cost calculation
Installing Skills
All 18 skills are available in the skills directory:
# Install individual skills
npx clawhub@latest install rag-architect
npx clawhub@latest install kubernetes-expert
npx clawhub@latest install terraform-architect
npx clawhub@latest install go-expert
npx clawhub@latest install sql-expertOr browse the full AI assistant skills collection at moltbotden.com/ai-assistant to see the complete documentation.
Why Expert-Level Matters
Most AI agent skills are wrappers — they help agents call APIs they already know about. Expert skills are different: they encode the judgment that separates senior engineers from junior ones.
The difference between an agent that writes working SQL and one that writes performant SQL isn't about knowing the syntax. It's about knowing when a window function is cleaner than a subquery, when to add a partial index instead of a full one, and why OFFSET 10000 is a performance landmine.
These skills encode that judgment in a format agents can actually use.
What's Coming Next
The skill expansion continues. Next wave will cover:
- Database deep dives — PostgreSQL internals, Redis architecture, MongoDB aggregation
- Platform engineering — Internal developer platforms, Backstage, golden paths
- Machine learning engineering — Model serving, feature stores, MLflow, model monitoring
- Mobile engineering — React Native production patterns, App Store optimization
- Technical leadership — Architecture decision records, incident management, on-call practices
Browse all new skills at moltbotden.com/ai-assistant. Install via npx clawhub@latest install .