By jadecli
Knowledge SDK with llms.txt crawler, doc fetcher, research orchestrator, and OTel tracker
npx claudepluginhub jadecli/claude-knowledge-sdk-typescriptSet up and manage OpenTelemetry monitoring for Claude Code sessions. USE THIS SKILL whenever the user asks about monitoring, telemetry, cost tracking, token usage, OTel configuration, Prometheus setup, or session analytics. Also trigger for "how much did that cost", "track my usage", "set up monitoring", "configure telemetry", or any reference to OTEL, Prometheus, Grafana, or the monitoring guide.
Orchestrate multi-agent research with recursive self-improvement. USE THIS SKILL whenever the user asks for comprehensive research, deep analysis, multi-source comparison, or investigation across documentation, codebases, and web sources. Trigger for: "research", "deep dive", "investigate", "compare", "survey", "comprehensive analysis", "find out everything about", or any question that needs multiple sources to answer well. This skill spawns parallel subagents, synthesizes findings, evaluates gaps, and recursively improves.
Fetch and index official Anthropic documentation for answering questions about Claude Code, the Agent SDK, MCP, skills, plugins, hooks, and monitoring. USE THIS SKILL whenever the user asks about Claude Code features, Agent SDK APIs, OTel configuration, cost tracking, secrets management, or any Anthropic platform documentation. Also trigger for "how does Claude Code...", "what's the SDK API for...", "check the docs", "latest changelog", or references to code.claude.com, platform.claude.com, or docs.anthropic.com. This skill acts as a local documentation oracle.
Parse and crawl llms.txt documentation indexes. Use this when you need to crawl docs, parse llms.txt, index documentation, fetch doc URLs, scrape docs, build a knowledge index, find what pages are in the docs, or work with llms.txt files from any site.
Convert structured task lists into Linear issues, cycles, and projects. Use when planning sprints, importing task lists, syncing progress to Linear, creating sprint cycles, or converting XML/markdown task lists into tracked issues.
Advanced Angular skills for dependency injection, RxJS, and signals.
Distilled knowledge from Anthropic's Claude ecosystem — as code, not docs.
A TypeScript SDK + Claude Code plugin that captures the patterns, types, and tooling from Claude Code v2.1.83, the Claude Agent SDK, claude-code-actions, and 8 Anthropic engineering blog posts. Includes 4 Claude Code skills, TypeScript LSP support, CI/CD workflows, and an llms.txt documentation crawler.
Instead of reading docs, you import them:
import {
runLoop, // Agent loop wrapping SDK query()
orchestrateResearch, // Lead + parallel subagent fan-out/fan-in
recursiveResearch, // Research → evaluate → improve loop
parseLlmsTxt, // Parse llms.txt documentation indexes
fetchAllKnowledge, // Fetch docs from 3 Anthropic surfaces
calculateBudget, // Context window management
generateOtelEnvVars, // OTel configuration generator
MODEL_PRICING, // Current model pricing table
} from '@jadecli/claude-knowledge-sdk';
# Install
npm install @jadecli/claude-knowledge-sdk
# Fetch documentation index
npx ck fetch-docs --priority critical
# Run a recursive research query
npx ck research "how does Claude Code handle context compaction"
# Generate OTel monitoring config
npx ck otel-setup --backend prometheus
This repo is also a Claude Code plugin with 4 skills and TypeScript LSP:
# Load as plugin (dev)
claude --plugin-dir ~/repos/claude-knowledge-sdk-typescript
# Reload after changes
/reload-plugins
# Use skills
/claude-knowledge-sdk:doc-fetcher
/claude-knowledge-sdk:research-loop
/claude-knowledge-sdk:otel-tracker
/claude-knowledge-sdk:llms-txt-crawler
| Skill | Description |
|---|---|
| doc-fetcher | Local documentation oracle — fetches and caches Anthropic docs |
| research-loop | Multi-agent recursive research with lead + subagent pattern |
| otel-tracker | OTel monitoring setup generator for Prometheus/SigNoz/Grafana |
| llms-txt-crawler | Parses llms.txt files, crawls doc URLs, generates Scrapy spiders |
The llms-txt-crawler skill can:
~/.claude/knowledge/Known endpoints: code.claude.com/docs/llms.txt, platform.claude.com/llms.txt
Built from 5 distilled sources:
| Source | What It Provides |
|---|---|
anthropics/claude-code v2.1.83 | CLI surface, hooks, skills, plugins, monitoring |
@anthropic-ai/claude-agent-sdk v0.2.33 | query(), AgentDefinition, subagents, V2 sessions |
anthropics/claude-code-action | CI/CD integration, GitHub Actions patterns |
anthropics/claude-code-monitoring-guide | OTel + Prometheus stack |
anthropic.com/engineering (8 posts) | Multi-agent patterns, context engineering, tool design |
Three GitHub Actions workflows are included:
ci.yml)Runs on push to main and PRs: typecheck → build → test → lint. All must pass.
claude-review.yml)Uses anthropics/claude-code-action@v1 for automated PR review.
security.yml)Uses anthropics/claude-code-security-review@main for AI security scanning.
| Secret | Source | Used By |
|---|---|---|
CLAUDE_CODE_OAUTH_TOKEN | Run claude setup-token locally | claude-review.yml |
CLAUDE_API_KEY | Anthropic API key with Claude Code usage enabled | security.yml |
Prerequisites:
/install-github-app in Claude Code)Follows Boris Cherny's "Programming TypeScript" discipline:
AgentId with SessionIdResult<T, E> replaces try/catch — no exceptions cross boundariesassertNever()noUncheckedIndexedAccess catches undefined array/object accessreadonly properties