tjdwls101010
UserCreate, validate, and maintain project-specific Claude Code harnesses through a structured interview.
Categories
Indexed Skills (7)
harness-creator
Design, generate, validate, and maintain a complete Claude Code harness (CLAUDE.md, rules, skills, hooks, permissions, agents, workflows) for a project through a structured interview. Use when the user wants to create or set up a harness / CLAUDE.md / skills / hooks for a project, improve or extend an existing .claude/ setup, or asks how Claude should be configured to work on their codebase. Also triggers on Korean requests like "하네스 만들어줘", "하네스 구성해줘", "클로드 세팅해줘".
market-scan
Analyze the US market regime, breadth, sector or industry strength, leadership, stock discovery, screening results, and watchlist candidates. Use whenever the user asks how the market is, which areas are strong, what tickers look promising, or to build or refresh a watchlist, even if Minervini is not named. Do not use for a buy, sell, hold, entry condition, or diagnosis judgment about one or a few named tickers; use ticker-analysis. Crypto, non-US listings, completed-trade grading, account-level position sizing, and capital allocation advice are outside scope.
ticker-analysis
Analyze one or a few named US-listed stocks for a prospective buy, entry timing, setup diagnosis, active-position HOLD or SELL evidence, re-entry, earnings risk, chart condition, or head-to-head comparison. Use whenever the user asks what to do with a specific US-listed ticker, even if Minervini or SEPA is not named. Do not use for market-wide sector screening or leader discovery; use market-scan. Crypto, non-US listings, completed-trade grading, account-level position sizing, and capital allocation advice are outside scope.
deploy-guard
Guard the deploy script against being pointed at production by accident. Use when the user asks to deploy, ship, or release, and when a deploy command is being drafted. Not for local builds or test runs, which no guard covers.
packaged
Fixture skill for the package-closure check. Use when exercising validate_harness.py against a plugin-packaged skill.
codex
Run OpenAI Codex (the `codex` CLI, GPT models) as a managed subagent from Claude Code — start work in the background and get a run handle back immediately, watch its live event log at a controlled level of detail, interrupt and redirect a run, resume any earlier Codex thread including ones started in the Codex TUI, and collect results as text or schema-validated JSON. Use whenever work is being handed to Codex or GPT, when a second independent model should look at something, when an external agent should keep working while Claude does something else, or when an earlier Codex session needs continuing. Also covers Codex CLI setup, auth, sandbox and config diagnosis. Triggers include: codex, 코덱스, 코덱스로, 코덱스한테, ���덱스에게, GPT, GPT한테, GPT에게 시켜, delegate to codex, run with codex, ask GPT, resume codex. Not for Claude's own subagents, the Task tool, or background Bash — those are Claude doing the work itself. Not for Codex Cloud or `codex mcp-server` / `app-server`.
ultra-fetch
Research a topic from web sources and secure the originals — fetch, crawl or map pages into clean, context-efficient markdown saved to a file, using a stealth browser that reaches sites the built-in WebFetch can't (bot-protected, Cloudflare, JS-rendered). This is the default way to read a web page's content, starting with a single page whose URL you already have, and extending to opening what a WebSearch turned up, reading across a whole site, or discovering what URLs a site has. Also the recovery when WebFetch hands back a refusal instead of a page — it reports a block as ordinary body text on a call it marks successful, and Korean news sites (yna.co.kr, mk.co.kr, fnnews.com) refuse it outright. NOT for a trivial quick fact where WebFetch already suffices, NOT for anything behind a login — it carries no credentials and no session, so authenticated pages, private feeds and paywalled articles are out of reach — and NOT for developing the ultra-fetch tool itself, which is ordinary repo work.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.