stock-direction-signals

Directional research on one US equity ticker or index proxy (MAG7, semiconductor and AI-infrastructure names, tokenized-stock watchlists, SPY/QQQ) for an agent on a MetaMask Agent Wallet. It runs a declarative manifest of nine paid reads from a fixed provider filter — market quote, daily chart, RSI, MACD, news sentiment, earnings history, Twitter/X chatter, Reddit threads, and macro regime data — and the agent fuses them into one brief: verdict (bullish / bearish / mixed / insufficient data), confidence, the technical setup, catalysts, social sentiment, macro backdrop, and what would invalidate the read. Research only — it never places orders and its output is never financial advice.

It wraps the stock-direction-signals manifest from SELAT-AI/selat-metamask-skills, and like selat-capabilities-for-metamask it needs no selat CLI and no SELAT plugin — the engine is the MetaMask Agent Wallet CLI (mm) plus the upstream repo's stdlib-only scripts. SELAT is the catalog and payment router, used keylessly; custody stays with the wallet.

Cost — read this first

• A run is a pipeline, not a menu: nine paid purchases, not one. Always dry-run first and show the user the live quoted total before executing.
• Every read is real USDC from the user's own self-custodial MetaMask Agent Wallet and its Circle Gateway balance. SELAT never holds keys or funds; every signature goes through the mm CLI.
• Prices live in the live 402 quotes and the manifest's caps, not here — the dry run prints the real quoted total; don't restate figures from memory.
• The manifest carries its own full-run spending cap, enforced by the runner; the Gateway deposit is the hard wallet-level cap above that.
• Non-advisory, always. The brief can support a bullish or bearish thesis but is research with sources, never advice, a prediction, or a signal to execute. If the user asks to trade on the result, decline that part — this skill buys data only.
• Never ask for, paste, or handle a private key.

Step 0 — get the tooling (free, no funds)

No selat CLI needed. Install the MetaMask Agent Wallet CLI if mm isn't on PATH, then fetch the upstream skills (text plus stdlib-only scripts):

mm --version || npm install -g @metamask/agent-wallet@latest
npx skills add SELAT-AI/selat-metamask-skills

Installing creates nothing money-related. The upstream repo is a prototype from a live 2026-08 integration spike — proven on mainnet, not yet hardened; review before large budgets.

Step 1 — dry run first, before any wallet setup

Do this before authenticating a wallet or asking the user to fund anything. The dry run quotes all nine steps for free — nothing is signed:

python3 skills/selat-purchasing/scripts/run_manifest.py \
  skills/stock-direction-signals/manifest.json \
  --param ticker=NVDA --dry-run

Show the user the quoted per-step prices and the total, and get their OK before wallet setup. If they decline, stop — nothing has been spent.

Step 2 — wallet setup (only after the user opts in)

mm doctor
python3 skills/selat-purchasing/scripts/eco_fund.py <usdc-amount>

Funding gotcha: each mm sign-in method (Google, email, MetaMask Mobile QR) loads a different wallet address — fund the wallet of the sign-in method the agent actually uses, and keep using that method. Wallet-escalated approvals arrive on that method's channel (email link, or push for Mobile QR); an approval that lands after a quote expires must be discarded and re-quoted.

Step 3 — run

python3 skills/selat-purchasing/scripts/run_manifest.py \
  skills/stock-direction-signals/manifest.json \
  --param ticker=NVDA \
  --param company=NVIDIA \
  --param "twitter_query=\$NVDA OR NVIDIA OR Blackwell" \
  --param "reddit_query=NVDA NVIDIA Blackwell stock" \
  --yes

Every step returns raw JSON — the synthesis is your job: score social tone, intensity, and engagement yourself (the reads return raw posts, not sentiment scores), weight Reddit for recency and cross-community repetition, then write the brief — verdict, confidence, technicals, catalysts, social sentiment, macro regime, contrarian risks, and what would invalidate the call. Relay it in plain language with what the run actually cost (python3 skills/selat-purchasing/scripts/spend_report.py); keep raw JSON and merchant URLs out of it. For multiple tickers, dry-run and run per ticker.

Why this is safe to install

• This wrapper is text-only — installing it writes one markdown file. The manifest it runs is declarative JSON; the runner scripts are Python standard library only and never touch key material.
• The wallet stays the authority: every signature and transaction goes through the mm CLI, and payments are quote-pinned through the SELAT router (the EIP-712 domain is always the router's, never a merchant's).
• Spend is triple-capped: per-step and full-run caps in the manifest, enforced by the runner, under the Gateway deposit as the hard wallet cap. A refused or expired quote charges nothing.
• The manifest pins a fixed provider filter — the agent must not substitute other catalog merchants into this skill.

Beyond this skill

For open-ended capability buying on the same wallet and tooling (web search, scraping, enrichment, perp funding rates, prediction-market data), use selat-capabilities-for-metamask. The other wrappers in this repo run vetted selat-skills recipes via the selat CLI — a vetted selat-skills variant of this same skill exists there too (SELAT-AI/selat-skills#64) for agents on a Circle Agent Wallet.

Docs: https://github.com/SELAT-AI/selat-metamask-skills