4 eli-alvarado

watchlist-manager Skill

管理美股/ETF 自选清单文件——添加、删除、更新、列出和校验股票代码,包含买入逻辑、入场价、目标价、止损、标签和每只股票的提醒规则。当用户说"把 NVDA 加进自选"、"给 AAPL 设 210 止损"、"我的自选里有什么"、"移除 TSLA"或"给这些打上半导体标签"时使用。

安装方式:把技能目录放入 ~/.claude/skills/(Claude Code)或在 claude.ai 设置中启用;也可复制右侧安装命令一键添加。

查看源码

技能指令原文(SKILL.md)

Watchlist Manager (US markets)

Maintains a single source-of-truth watchlist file that the price-alerts and watchlist-brief skills read.
Informational tool only. Never place, modify, or cancel a trade. Editing the watchlist is not trading.

File location

Shared with price-alerts and watchlist-brief. Survives npx skills add reinstall. Do not cd into this skill folder to run scripts (that folder is replaced on update). Run from the project root by path, e.g. python3 .claude/skills/watchlist-manager/scripts/watchlist.py init.

  1. $WATCHLIST_FILE if set.
  2. Else ./watchlist.json in the project directory. If cwd is inside .claude/skills/ or .agents/skills/ (this folder, or scripts/), walk up to the project root and use that watchlist.json so the skill folder and project root cannot diverge. YAML watchlist.yaml works if PyYAML is installed and you pass --file.
  3. Else ~/.config/us-watchlist-alerts/watchlist.json (Windows: %APPDATA%\us-watchlist-alerts\watchlist.json; or $XDG_CONFIG_HOME/us-watchlist-alerts/).
  4. A leftover watchlist.json inside a skill folder is copied to that destination on first run so reinstall cannot delete it.
  5. If none exists, init writes an empty valid list at that default path (init / add / list print using ). Ask before overwriting an existing file.

Schema (version 1)

{
  "version": 1,
  "settings": { "cooldown_minutes": 240, "rearm_pct": 0.5, "sessions": ["regular"],
                "volume_spike_multiple": 2.0, "near_52w_pct": 2.0, "earnings_days": 3,
                "max_alerts_per_run": 20, "extra_closures": [], "stale_after_seconds": 900 },
  "tickers": [{
    "symbol": "NVDA",            // required, uppercase; class shares as BRK.B or BRK-B
    "side": "long",              // long | short | watch
    "added": "2026-10-15",
    "thesis": "AI datacenter capex",
    "entry": 160.0, "targets": [190.0, 220.0], "stop": 140.0,
    "tags": ["ai", "semis"],
    "alerts": [ {"type": "price_below", "value": 150, "cooldown_minutes": 60} ],
    "notes": ""
  }]
}

Alert type values and their value meaning are documented in ../price-alerts/SKILL.md (sibling skill after install).
A stop implies a stop_hit alert and targets imply target_hit automatically.

Operations (prefer the script; it validates and writes atomically)

scripts/ ships next to this SKILL.md after npx skills add. Run from the project root by path (or from a git clone; root scripts/ is the same files):

python3 scripts/watchlist.py init
python3 scripts/watchlist.py init --example
python3 scripts/watchlist.py init --list-presets
python3 scripts/watchlist.py init --preset mega-cap-tech
python3 scripts/watchlist.py add NVDA --thesis "AI capex" --entry 160 --target 190 --target 220 --stop 140 --tag ai --tag semis --alert price_below:150
python3 scripts/watchlist.py add NVDA --file swing.json --alert price_below:150
python3 scripts/watchlist.py add NVDA --no-verify
python3 scripts/watchlist.py update NVDA --stop 145
python3 scripts/watchlist.py update NVDA --tag megacap --alert rsi_above:75 --append
python3 scripts/watchlist.py remove NVDA
python3 scripts/watchlist.py list [--tag semis]
python3 scripts/watchlist.py show NVDA
python3 scripts/watchlist.py validate

add looks the symbol up on the Yahoo chart (same default quote path as quotes.py). Unknown tickers are rejected (NVDA is not a known US ticker…). --no-verify / WATCHLIST_NO_VERIFY=1 skips the lookup when you are offline. If the chart request fails because the network is down, the CLI prints a WARN and adds the ticker anyway. --fixture / WATCHLIST_FIXTURE checks the bundled sample quotes instead — presets work in that mode (the fixture _default alias covers built-in preset symbols that are not in the 3-name demo).
If Python is unavailable, edit the JSON by hand following the schema and the validation rules below.

Workflow

  1. Resolve the ticker. Uppercase it. If the user gives a company name, confirm the symbol (e.g. "Alphabet" -> GOOGL or GOOG?). Reject non-US listings or say they are unsupported by default data.
  2. Sanity-check with a live quote: add already checks the symbol on the Yahoo chart (skip with --no-verify). Run python3 scripts/quotes.py NVDA when you want the price printed. If the user's stop is above price for a long (or target below), point it out and ask.
  3. Capture the why. If no thesis is given, ask once for a one-line reason; store "" if they decline. Don't invent a thesis.
  4. Suggest, don't impose, alerts. Reasonable defaults to offer: stop/target levels, earnings_within: 3, pct_move: 5 (last vs prior close). If they want a drop that trades through a % and then recovers, offer opt-in intraday_low_pct / intraday_extreme_pct and say those use session low/high, not last. Mention iv_above/intraday_range_pct only if they ask about vol or today's range — ATM IV is unofficial Yahoo, not IV rank. Only add what they agree to.
  5. Write, validate, confirm. Show the resulting entry (symbol, entry, stop, targets, alerts) in one compact line.

Validation rules

  • Symbol matches ^[A-Z]{1,5}([.-][A-Z]{1,2})?$; no duplicates.
  • add also checks the symbol on the Yahoo chart unless --no-verify is set. Reject unknown symbols. On network outage, warn and add anyway.
  • Built-in presets (mega-cap-tech, index-etfs, sector-etfs, dividend-staples) are valid symbol lists; init --list-presets prints them.
  • entry, stop, targets[] are numbers. For side: long, stop < entry; for side: short, stop > entry.
  • settings.sessions if present is a non-empty list of regular / pre / after.
  • settings.extra_closures if present is ISO dates (YYYY-MM-DD) for unscheduled full-day closures.
  • Every alert type is a known type.

Listing format

Compact table: Ticker | Side | Entry | Stop | Targets | Tags | Thesis. Add a live price column only if the user asks or you're already fetching quotes.

Don'ts

  • Don't delete tickers without explicit instruction; for "clean up my list" propose removals first.
  • Don't give buy/sell recommendations. You may describe levels and risk (e.g. "stop is 12% below entry").