ai-native-camp/camp-2
12 skills14.5k installs204 starsGitHub
Install
npx skills add https://github.com/ai-native-camp/camp-2Skills in this repo
1Day1 OnboardingDay1 Onboarding is a structured agent skill from an AI-native camp curriculum that walks new Claude Code users through day-one environment setup after README Step 0 installation. It combines explain-then-execute pedagogy: official documentation links, optional editor selection without pressure to leave the terminal, Mac-specific Finder visibility for dotfolders, and configuring Explanatory output style so the agent narrates reasoning while coding. Participants run `claude` for Anthropic login and subscription validation, then open the project root to see camp materials under .claude/skills. Solo builders adopting Claude Code for the first time can reuse the same checklist even outside the camp to reduce friction before skills, MCP, or repo work. The content is Korean-forward with clear tables and commands suitable for guided facilitation by an agent coach.1.4kinstalls2Day3 Clarifyday3-clarify is a curriculum skill from AI Native Camp that trains solo builders to run a Clarify pass before Claude Code Plan or implementation. It explains why Plan alone cannot fix vague input: Plan spends tokens on how to build without first deciding what success looks like. The skill walks through failure modes (e.g. “make a logo” with no brand, palette, or surface) and shows how a short clarification turn produces concrete targets for Plan. It anchors the four-step CC Workflow—clarify, plan, implement, wrap—as the minimum serious agent loop. For Prism builders using Claude Code, Cursor, or similar agents, this is procedural knowledge for any creative or feature request: invoke when the request is underspecified, when you would otherwise enter Plan immediately, or when you keep re-explaining the same correction. It is methodology, not an integration; no external APIs. Confidence is high for Claude Code workflows; adapt AskUserQuestion to your client’s equivalent structured-question tool.1.3kinstalls3Day4 Wrap And AnalyzeDay4 Wrap and Analyze is a camp-2 agent skill that teaches solo builders how to close a Claude Code session with a structured multi-agent pipeline instead of one monolithic recap. Phase 1 spins up four parallel sub-agents to flag doc updates, spot automation opportunities, extract learnings, and suggest follow-ups; Phase 2 runs a duplicate-checker to merge overlapping recommendations. The narrative frames this as an end-of-day standup and extends earlier single-subagent Task usage into true parallel delegation. It is instructional workflow knowledge for Claude Code rather than a domain-specific integrator. Use it when you want session-wrap discipline, portfolio-style agent design practice, or a template for operating rituals before you iterate the next day.1.3kinstalls4Day1 Test Skillday1-test-skill is a Camp 2 onboarding smoke test for AI-native builders learning Claude’s skill system. When you invoke /day1-test-skill, the agent reads `.claude/skills/day1-test-skill/SKILL.md` and prints a prescribed success message that walks through what just happened: command in, markdown recipe loaded, structured output out. It deliberately contrasts skills with CLAUDE.md by noting skills load only when called. For solo builders new to agentic workflows, this is the fastest proof that directory layout, naming, and triggers work before you author real build or ship skills. It is not a testing framework for application code—it is meta validation of your skill installation.1.3kinstalls5Day2 Mcp And Context SyncDay2 MCP and Context Sync is a camp-2 agent skill that grounds solo builders in the Model Context Protocol before they plug real tools into Claude Code. It defines Host (the AI app), Client (connection manager), and Server (tool provider), and shows why MCP beats one-off Slack, Notion, and calendar integrations. The block opens with required official documentation links for Claude Code MCP and the MCP introduction site, then progresses from concept to context synchronization so agents share a coherent project picture. It is foundational integration literacy for indie agent stacks, not a specific vendor server implementation. Use when you are onboarding MCP, debugging why tools do not appear, or aligning camp day-2 exercises with production-style agent architecture.1.3kinstalls6History InsightHistory Insight is an agent skill that teaches solo builders how Claude Code persists sessions in .jsonl files and how to strip noise before analysis. Most file volume is file-history-snapshot and queue-operation metadata; only user lines and assistant text blocks carry the dialogue you care about for retros, prompts tuning, or compliance review. The skill catalogs each type field, shows proportional size from a real large session example, and supplies jq filters to emit compact conversation streams. Use it when exports are too heavy for git, when you want accurate token or topic stats, or when an agent must summarize past work without loading megabytes of tool traces. It is documentation-forward meta guidance—not a hosted parser service—so you still run jq locally. Helpful across Build agent-tooling and Operate iterate when debugging long sessions.1.2kinstalls7Session WrapSession-wrap packages multi-agent orchestration patterns for Claude Code: how to mirror a task’s dependency graph in agent architecture so independent work runs in parallel and chained work stays sequential. Solo builders use it when one agent session grows into research plus implementation plus review, or when they need fan-out analysis without garbled merged answers. The skill encodes Anthropic-oriented principles—routing by input type, orchestrator-worker for dynamic coding tasks, and evaluator-optimizer loops when quality matters—and contrasts them with naive single-thread prompting. It is editorial guidance and pattern vocabulary, not a hosted orchestrator or MCP server. Install it when you are designing agent roles, spawn rules, and handoffs in Camp-style or custom workflows, especially before committing to a multi-subagent layout that will be expensive to unwind.1.2kinstalls8Fetch TweetFetch Tweet is a small integration skill for solo builders who work in Korean or English with X/Twitter links in the agent chat. When you paste a status URL or ask to read or translate a post, the agent runs fetch_tweet.py, which rewrites the host to FxEmbed’s API and returns structured tweet data. That makes it easy to quote founders, compare launch announcements, or feed tweet text into summarization without manual copy-paste. Other camp skills can depend on the same script for programmatic access via --json. Network access is required to reach api.fxtwitter.com; there is no OAuth or official X API setup. Tag it as multi-phase because the same fetch supports Idea research, Grow content monitoring, and Launch distribution reconnaissance, while the implementation itself lives under Build integrations.1.1kinstalls9Content DigestContent Digest is an agent workflow for solo builders who learn from scattered links and documents instead of starting from scratch. When you paste a YouTube, X/Twitter, webpage, or PDF URL—or ask to digest or summarize in Korean or English—the skill detects the type, delegates heavy extraction to a Task agent, and lands a clean markdown digest you can reread without bloating the main chat. Its Quiz-First design front-loads nine questions across three difficulty levels, leaning on pretesting and knowledge-gap psychology so you retain more than passive summaries. Architecture rules keep transcripts clean (no timestamps clutter), integrate keyword web research, and separate context so your primary session only consumes the finished file. You choose how deep to go after the quiz, including foundational concept expansion beyond the original piece. It pairs naturally with research in early journey phases and with content learning later, but it is not a generic note app or automated course platform.1.1kinstalls10Session AnalyzerSession Analyzer is a reference skill for solo builders who ship with Claude Code and need to reconstruct what an agent actually did. Instead of guessing from chat summaries, you apply repeatable grep and search patterns against per-session debug logs under ~/.claude/debug/. The skill catalogs how SubagentStart and SubagentStop queries appear, how PreToolUse hooks fire before tools like Write, and how hook matchers report matched counts before deduplication. It also covers subagent session registration and frontmatter hooks loaded from child agents. Use it when a multi-agent task mis-routed, a hook blocked a tool, or you need evidence for support and iteration. It does not replace structured telemetry products; it gives you fast, local forensics without leaving the terminal. Pair it with your own notes on which hooks you enabled so patterns map back to your repo’s agent configuration.1.1kinstalls11Team Assembleteam-assemble is an agent orchestration skill for solo builders who need more than one specialized pass on a hard problem. Instead of one long chat, you describe the goal—such as migrating Supabase to PostgreSQL or producing a competitive analysis report—and the skill proposes a named team with roles, model choices, responsibilities, and dependency columns. Phase two materializes that plan using team and task primitives (TeamCreate, TaskCreate, TaskUpdate with blocked-by links). Phase three runs execution in rounds: independent tasks first, then tasks that need prior outputs, with explicit parallelism where safe. The pattern fits whenever work naturally splits into analyst, planner, implementer, and validator roles. It does not replace domain skills for SQL or research; it sequences them so your coding agent behaves like a small staff. Review tool permissions and your environment’s support for multi-task APIs before relying on it in production repos.1.1kinstalls12CompoundCompound is a journey-wide agent skill that defines a strict markdown schema for turning one-off wins and failures into reusable insight documents. Solo builders install it when an AI coding session surfaces a lesson worth keeping—not a README tweak, but a structured record with typed frontmatter, narrative sections that contrast failed attempts with what actually worked, and an explicit replication guide for the next time a similar situation appears. The skill also standardizes pattern references as anti-pattern versus best-practice triplets with structural rationale and pointers into a knowledge folder, which helps agents avoid repeating known mistakes during Build, Ship, Launch, or Operate work. It fits indie workflows where the same agent wears every hat: you document once in compound format, tag impact, and let later invokes pull the right pattern by insight_type or domain without re-deriving context from chat history.1.1kinstalls