Price Alert Skill

Parameterized price-alert management. The user can set, list, and cancel alerts on any ticker; they fire a Telegram push when the threshold hits.

Architecture

price-alert/
├── SKILL.md              # this file
├── alerts.json           # active alert definitions (committed to git)
├── alerts_fired.log      # append-only history of triggered alerts
├── scripts/
│   ├── add_alert.py      # add new alert
│   ├── list_alerts.py    # list all / active / fired
│   ├── cancel_alert.py   # cancel by id, ticker, or --all
│   ├── check_alerts.py   # run by GitHub Actions cron, sends Telegram
│   └── chat_handler.py   # Option A chat path: poll Telegram + tool-use via Claude API
└── webhook/              # OPTIONAL Option B chat path (Cloudflare Worker)
    ├── worker.ts         # TypeScript webhook handler (mirrors chat_handler.py)
    └── wrangler.toml     # CF Worker deploy config

Price scanning runs in GitHub Actions (.github/workflows/price-alerts.yml), which executes check_alerts.py every 2 minutes 24/7 and pushes Telegram notifications for any triggered alerts.

The bidirectional NL chat path has two interchangeable implementations — pick one per the latency you want:

• Option A (default): telegram-chat.yml GH Actions cron polls Telegram every 2-5 min, runs chat_handler.py. Latency 2-15 min. $0. Setup: SETUP.md.
• Option B (optional upgrade): Cloudflare Worker (webhook/worker.ts) receives Telegram webhook POSTs and processes instantly. Latency 1-3 sec. Still $0 (CF free tier 100k req/day). Setup: SETUP-WEBHOOK.md on top of basic SETUP.

Both paths use the same Anthropic API and end up modifying the same alerts.json via GitHub. They are not used together — Option B replaces Option A entirely once you enable it.

Decision tree

After running any add/cancel, commit and push alerts.json so GitHub Actions sees the change on its next run.

Adding alerts — parameter mapping

# Absolute price thresholds
python add_alert.py TICKER below PRICE [--note "..."]
python add_alert.py TICKER above PRICE [--note "..."]

# Percentage moves (auto-anchors to current price unless --anchor specified)
python add_alert.py TICKER drop PCT  [--note "..."] [--anchor PRICE]
python add_alert.py TICKER rise PCT  [--note "..."] [--anchor PRICE]

Always include --note when context is meaningful — it shows up in the Telegram message and helps you remember why you set it weeks later.

Examples

# AI glass substrate retracement watchlist
python add_alert.py GLW below 140 --note "AI glass tier 1 entry"
python add_alert.py ONTO below 210 --note "Inspection tools post-reset"
python add_alert.py LRCX below 240 --note "50DMA support"
python add_alert.py KLAC below 1500 --note "Conservative semicap entry"

# Macro flip warnings
python add_alert.py SPY drop 5 --note "Material pullback from current ATH"
python add_alert.py SMH drop 10 --note "Semi correction trigger"

# Take-profit alerts on long positions
python add_alert.py NVDA above 1300 --note "Target reached - consider trim"

Telegram message format

When an alert fires, the user receives:

🔻 PRICE ALERT: GLW

Current: $138.45
Trigger: $138.45 ≤ $140.00
52W high: $198.25 (-30.2% off)
52W low: $46.34

Note: AI glass tier 1 entry

Alert id: glw-below-a3b2c1
Set: 2026-05-10

GitHub Actions setup (one-time)

Required GitHub repo secrets:

• TELEGRAM_BOT_TOKEN — from @BotFather
• TELEGRAM_CHAT_ID — from https://api.telegram.org/bot<TOKEN>/getUpdates

Workflow lives at .github/workflows/price-alerts.yml. Cron is */15 13-21 * * 1-5 (every 15 min during US trading hours, weekdays).

To trigger manually for testing: GitHub repo → Actions tab → "Price Alert Checker" → "Run workflow".

Important behaviors

• Idempotent: once fired, an alert is marked fired=true and won't re-trigger. To get the same alert again, run cancel_alert.py <id> --rearm.
• Anchors are static: drop 10% from anchor $100 fires at $90 forever — even if the stock first drops to $85 and then rebounds to $95 before falling to $90 again. The first $90 cross fires it.
• GitHub Actions auto-commits the updated alerts.json when alerts fire, keeping the source of truth in git.
• No re-fire: after firing once, the alert stays in alerts.json (so you can audit history) but is filtered out of list_alerts.py --active.

When NOT to use this skill

• Intraday scalping: 2-min granularity is too coarse. Use a real broker alert.
• Composite conditions (e.g., "GLW < 140 AND VIX > 25"): not supported by current schema. Set two alerts and use your judgment.
• Crypto / non-US tickers: not tested. yfinance support varies.

⚠️ Alert fired — what to do BEFORE adding

A fired alert is a trigger to research, not a buy signal. When the user receives an alert and asks "should I add now":

1. Run `analyze-stock TICKER` — has the bull thesis changed since the alert was set? Fresh insider check, valuation, catalysts.
2. Trigger `macro-warning` — if regime flipped to 🔴 RED between alert-set and alert-fire, the entry price you wanted is no longer the entry price you want.
3. Verify the 3-tier plan — the alert price was T1. If T1 hit because of macro selloff (not stock-specific), T2 and T3 likely follow. Pre-commit to tier sizes now.
4. Position-size check — if adding pushes the position above 10% of book, scale the add down.
5. Insider strict 30d — insider_ratio.py TICKER --window 30 — catch any pre-alert C-suite distribution that explains the drop.

"Look carefully" rule: alerts fire because of price moves; price moves often have a REASON. Don't add without finding the reason. If you can't articulate why the stock dropped to your alert price, you don't yet understand whether the bull thesis is intact.

Verification

To verify GitHub Actions is working without waiting for a real trigger:

1. Set a guaranteed-fire alert: add_alert.py SPY below 99999
2. Manually trigger workflow in GitHub UI
3. Check Telegram for message
4. Cancel: cancel_alert.py <id>