robinhood

Imported from bottlefedchaney808/BlackHole (.claude/skills/robinhood/SKILL.md). Install upstream with npx skills add bottlefedchaney808/BlackHole --skill robinhood. Copyright stays with the author.

Overview

Robinhood connects as an MCP server (mcp__robinhood__*, ~50 tools, deferred — load via ToolSearch("select:mcp__robinhood__...") before first use). This is a live brokerage account, not a sandbox — place_equity_order/place_option_order/exercise_option/cancel_* are real, irreversible-in-spirit actions. Always confirm with the user before submitting or canceling an order; review_equity_order/review_option_order (dry-run risk/cost preview) is safe to call freely and should be used before every place_* call.

Almost every tool response includes a guide field with binding formatting/behavior rules (masking account numbers, which field is authoritative for buying power, sort order, etc.) — those rules live in the tool response itself, not duplicated here, since they can change server-side. Follow them.

Accounts (inventory as of 2026-08-18)

Two individual margin/limited-margin accounts, both option_level_3. Numbers are masked to the last 4 digits because this repo is public: get_accounts returns the full numbers, and the agentic one is ROBINHOOD_AGENTIC_ACCOUNT in the root .env. Mask to ••••XXXX in any user-facing text too, per the get_accounts guide.

Only account B (••••1659, "Agentic") is agentic_allowed: true — that's the one this agent can place/cancel orders in. Account A is visible (read) but not tradable by this agent; if the user wants a trade in A, say it's not accessible to this agent, don't imply it needs enabling on Robinhood's side.

Snapshot at inventory time (stale the moment markets move — always re-fetch before making decisions, this is just orientation):

• Account A (••••8556): total value ~$53.88 (equity $6.27, options $37, cash $10.61). Positions: 0.028 sh NVDA (fractional, avg $211.94). Options: an SPY call spread (long/short $26/$23 avg, exp 2026-08-28), a long SPCX call (avg $13, exp 2026-10-16), an NVDA call spread (long $47 / short $37, exp 2026-11-20), a CRGY call spread (long $50 / short $16, exp 2026-09-18).
• Account B (••••1659, Agentic): total value ~$527.27 (equity $121.97, cash $405.30, $350 pending deposit). Positions: 3 sh UUUU (avg $14.17), 8 sh TGB (avg $8.86), 3 sh KOS (avg $2.57). No open options.

Re-run get_portfolio + get_equity_positions + get_option_positions(nonzero=true) per account before quoting current numbers — don't trust this table for anything but "what generally exists."

Watchlists

get_watchlists → 4 custom lists (all owner_type custom, writable):

• Options watchlist (5f643fd5-a094-48a8-ad22-1a0d8a756617) — single-leg option contracts, 14 items at inventory time but get_option_watchlist returned empty on the last check; re-poll, don't trust the stale count. Use get_option_watchlist, not get_watchlist_items, for this one (the latter 400s on option strategy items).
• Crypto to watch (63604e85-...) — 6 items.
• First list (ad481bfe-...) — 16 items, general/unlabeled.
• Sell Vol Plays (dad631c6-...) — 8 items, described as "Aug 5 post-earnings IV-crush sell-vol candidates (SEDG CVS IRM CPRI PRGO SHAK DK HMC)" — a vol-selling watchlist the user already curated; a natural target basket for a "trading mode" IV-crush/mean-reversion sweep.

Use get_watchlist_items(list_id) for the non-options lists.

Scanners (Legend screeners)

get_scans → 4 saved scans, notably two vol-relevant ones already built:

• "Volatility Comparison" (2c7667c4-..., Cortex-managed / read-only) — ATM IV(30d), HV(1M), IV rank, IV−HV delta, options volume, sorted by options volume desc. This is the fastest single call for "where's IV rich vs. realized right now" across the market.
• "High options volume and IV" (ad2715de-..., editable) — ATM IV(30d), options volume + relative options volume, sorted by IV desc. Good for unusual-activity-meets-elevated-IV screens.
• Two generic "Untitled Scan"/"Untitled Scan 2" (market-cap sorted, no vol columns) — not vol-specific, ignore for trading-mode work unless the user repurposes them.

Call get_scanner_filter_specs before constructing/editing any filter (create_scan, update_scan_filters) — don't guess filter_type_enum values. Cortex-managed scans can be run_scan'd but not modified; clone into a new scan via create_scan instead.

run_scan on a broad/unfiltered scan can exceed the tool-result token limit (seen live: 52,685/88,697 characters over the max) — the harness falls back to saving the output to a side file rather than returning it. Prefer scans with tighter filters, or check for a results-limit param, before running a scan you expect to return a large universe.

Tool map (by task)

• Account/portfolio: get_accounts, get_portfolio, get_limited_margin_upgrade_info, get_option_level_upgrade_info.
• Positions & history: get_equity_positions, get_option_positions, get_equity_tax_lots, get_pnl_trade_history, get_realized_pnl, get_equity_orders, get_option_orders.
• Quotes & chains: get_equity_quotes, get_equity_price_book, get_option_chains, get_option_quotes, get_option_instruments (strike/type lookup from an option_id), get_equity_historicals, get_option_historicals, get_index_quotes/get_index_historicals.
• Fundamentals/context: get_equity_fundamentals, get_financials, get_earnings_calendar, get_earnings_results, get_equity_technical_indicators, get_equity_tradability.
• Screeners: get_scans, run_scan, create_scan, update_scan_filters, update_scan_config, get_scanner_filter_specs.
• Watchlists: get_watchlists, get_watchlist_items, get_option_watchlist, create_watchlist, update_watchlist, add_to_watchlist/remove_from_watchlist, add_option_to_watchlist/remove_option_from_watchlist, follow_watchlist/unfollow_watchlist, get_popular_watchlists.
• Orders (live, act with care): review_equity_order → place_equity_order → cancel_equity_order; review_option_order → place_option_order → cancel_option_order; exercise_option, cancel_option_exercise.

Working conventions

• Never default an account_number from get_accounts when a write tool requires one explicitly per its schema (get_equity_positions, get_option_positions, etc. say "must come from the user or be clearly implied") — ask or infer from context, don't silently pick account A vs. B.
• Buying-power questions always route through get_portfolio, never derive from get_accounts.
• Before any place_option_order/place_equity_order, call the matching review_*_order first and surface the cost/risk preview to the user; wait for explicit go-ahead before placing.
• Claude Code auto-mode classifier blocks place_equity_order / place_option_order. Do not retry the same tool after "Blocked by classifier" / "Permission denied by Claude Code auto mode classifier". Tell the user auto-mode cannot place live orders.
• This account's options are multi-leg (spreads) in account A — get_option_watchlist and get_option_positions only show single-leg detail per contract; a spread shows as two separate long/short legs sharing a chain_id+near-identical opened_at, not as one strategy object. Group by chain_symbol + opened_at proximity when presenting a "position" to the user.
• get_equity_quotes/get_option_quotes's symbols param is a JSON array, not a comma-joined string — pass ["KOS","TGB"], never "KOS,TGB" (the latter 400s: "has type "string", want one of "null, array"").
• place_option_order rejects any top-level property outside its schema — e.g. underlying_type, chain_symbol — with "unexpected additional properties"; only pass the fields the schema defines. Re-verified 2026-09-03 against a live 400: multi-leg orders (2+ legs in one place_option_order call) are rejected for account B with "Multi-leg options orders aren't supported in Robinhood agentic accounts yet." — this is a Robinhood-side restriction on agentic accounts specifically, NOT gated by option_level (B is option_level_3, same as A). The tool's own schema/description doesn't surface this restriction — it only mentions option_level gating — so don't trust the schema text alone here; the 400 message is the ground truth. A straddle/strangle is still achievable by placing the call leg and the put leg as two independent single-leg orders (each reviewed and placed separately) — this gives the same economic position, just without atomic net-debit execution (one leg can fill without the other; check both fills before treating the position as fully on).