AI & LLM Integration for Claude Code

AI Operations Team — Embed: Embedding model selection, vector pipeline design, similarity search, and production index management.

Design embedding pipelines — model selection, batching, normalization, index refresh strategy.

Optimize similarity search — ANN index tuning, hybrid search, reranking, query expansion.

Design eval harnesses — task schemas, metrics, dataset versioning, eval-as-code patterns.

AI Operations Team — Evals: Eval harness design, benchmark suites, automated regression, human eval orchestration.

AI Operations Team — Guard: Input/output safety filters, PII detection, content moderation, and AI policy enforcement at runtime.

Design guardrail layers — input classifiers, output validators, PII scrubbers, policy rule engines.

Mesh skill: mesh-observe

Mesh skill: mesh-recon

AI Operations Team — Prompt: System prompt design, few-shot libraries, chain-of-thought patterns, and prompt versioning for production LLM applications.

Design production prompts — system prompt architecture, instruction clarity, few-shot selection.

Audit prompt library — duplication, quality, coverage gaps, version drift, eval alignment.

Build prompt versioning systems — storage, A/B testing, regression tracking, rollback.

AI Operations Team — Rank: Retrieval reranking, relevance scoring, learning-to-rank pipelines, and result quality evaluation for search and RAG.

Design ranking pipelines — reranker selection, score fusion, cross-encoder patterns, latency trade-offs.

Build ranking evaluation — NDCG/MRR measurement, human relevance labeling, offline eval harness.

Serv skill: serv-cold

Serv skill: serv-design

|

Terra skill: terra-drift

Terra skill: terra-recon

AI Operations Team — Token: Context window optimization, token counting, truncation strategies, and chunking patterns for LLM efficiency.

Design token budgets — system/user/assistant allocation, overflow handling, context compression.

Design chunking strategies — semantic splitting, overlap tuning, retrieval-aware chunk sizing.

AI Operations Team — Trace: LLM tracing, span capture, prompt/completion logging, cost attribution, and AI system debugging.

Debug AI system behavior using traces — prompt reconstruction, output comparison, failure attribution.

Instrument LLM calls with tracing — span structure, token counts, latency, model metadata.

Audit LLM observability coverage — trace gaps, logging completeness, cost attribution accuracy.

Infrastructure and tools for decentralized messaging protocol integration and messaging agent development.

Skill authoring workflow for creating and validating new agent skills.

Make your AI agent code with your project's architecture, rules, and decisions. Git-native context (folder conventions, ADRs, team standards) so new code lands in the right place and decisions persist across sessions and agents.

Real-time web search and data extraction for AI agents - powered by Nimble. Search, extract, map, and crawl any site on the web with built-in stealth, JS rendering, and retries. Get clean, structured web data from any public webpage with pre-built Web Search Agents, or build your own reusable extraction templates. The definitive web data toolkit for AI-native development.

The /new-dcode-agent skill: scaffold a Deep Agents agent or a dcode CLI agent from one command.

Analyze and reduce Claude Code token overhead

Skills and tools powered by the Honeydew MCP that help coding agents query data and build semantic models

Claude skill for Feishu (Lark) platform integration and workplace automation.

Essential AI agents: agent-manager, code-analyzer, file-analyzer, test-runner