trader-warroom

The War Room runs the other ten steps of a day trade for you, in order, and puts the answer on one page. It does no math of its own: every number comes from trader-entry, trader-exit, trader-timemachine, trader-budget, trader-risk and trader-news, run exactly as their own skills run them. What it adds is speed (a whole scan, every breakout replayed, in under a minute), one picture of all of it, and the one honest check none of them can do alone: on a replay, what the rest of the day really did. Most people who day trade lose money. A good-looking dashboard does not change that.

${CLAUDE_SKILL_DIR} below is this skill's folder (where this SKILL.md lives). It needs the other trader-* skill folders next to it (install the whole pack).

python3 "${CLAUDE_SKILL_DIR}/warroom.py" run TSLA                          # live, now
python3 "${CLAUDE_SKILL_DIR}/warroom.py" run TSLA --now "2026-10-22 10:15" # replay a past session
python3 "${CLAUDE_SKILL_DIR}/warroom.py" run TSLA --now "2026-10-22 10:15" --at-trigger
python3 "${CLAUDE_SKILL_DIR}/warroom.py" run TSLA --stop or-low --target 1R
python3 "${CLAUDE_SKILL_DIR}/warroom.py" scan                              # ~110 liquid names, the 8 latest breakouts
python3 "${CLAUDE_SKILL_DIR}/warroom.py" scan --now "2026-10-22 10:15" --at-trigger --top 10
python3 "${CLAUDE_SKILL_DIR}/warroom.py" scan --symbols TSLA,NVDA,AMD      # or --watchlist FILE, or --universe
python3 "${CLAUDE_SKILL_DIR}/warroom.py" pick NVDA                         # promote one ticker into trader/
python3 "${CLAUDE_SKILL_DIR}/warroom.py" open                              # open the last dashboard

Other flags: --replays N (10,000), --seed N (19), --or-minutes N (15), --top N (8; 0 replays every breakout), --min-price, --max-confirm, --no-news, --no-promote, --offline PATH (saved 60-day charts), --open (open the page when done), --json.

What it does, per ticker

Each ticker runs in its own sandbox, trader/warroom/<run>/<TICKER>/trader/, seeded with the user's trader/config.json (or the example config, said plainly), portfolio.json and today's broker log, so the risk check counts what was already placed today. Sixty days of 5-minute bars are fetched once and handed to every tool as --offline chart.json, so all of them see the same data at the same --now.

Step 4 matters. A breakout that fired at 9:50 and has run since is a worse trade at 10:15: the same stop is further away, so fewer shares fit, and the target is closer, so reward:risk falls. trader-budget and trader-risk already measure from the higher of the plan's entry and a fresh price; the war room just hands them the fresh price. A plan whose stop or target was already hit before --now is marked PLAYED OUT: it is history, not a trade.

Statuses

Replays and what really happened

--now in the past is a replay. Everything is judged with only the bars closed by then; the bars after it are drawn faded, and the orange line shows what the price really did with the plan: stop first, target first, or where it sat at the flat-by time, and where that close fell in the time machine's cone. --at-trigger judges each breakout at the moment it fired instead of at --now: a fair replay of what you could have taken. A scan replay ends with one line on how the day's plans did, and says that one day is one day.

A replay is research. It is never promoted and never becomes an order.

Scan

scan runs trader-scanner's opening-range scan (scan.py orb) on the liquid list (about 110 names), your --symbols or --watchlist, or --universe (the whole market from trader/universe.txt: minutes, not seconds), then replays the --top latest breakouts (closest to the price now), four at a time. The board ranks them by the replays' average R. That is a ranking of simulations, never a list of trades to take. Live, the scan's watchlist is also copied to trader/watchlist.json for trader-entry.

Live: from the board to the broker

A live run whose order passed trader-risk (and is not played out) is promoted: entry.json, plan.json, timemachine.json, orders.json and the fresh price are copied into trader/, and check.py runs again there against the real broker log. For a scan, warroom.py pick TICKER does the same for one row of today's live board. Then /trader-broker shows the ticket and asks the user yes or no. The war room itself never calls a broker tool.

The dashboard

trader/warroom/<run>/warroom.html, one self-contained page (no network, light and dark): the leaderboard with each breakout's odds bar and replay cone; then, per ticker, the session's candles with the opening range, the entry, stop and target, the cone of 10,000 futures with 60 sample paths, the odds, where the replays ended, every past session the rule fired (and today), the bracket ticket, the risk lines, the headlines, and the raw output of every tool it ran. Hover any chart for the numbers. Open it with warroom.py open or --open.

Steps

1. If trader/config.json does not exist, say the run will use the example limits ($5,000, $50 max loss a trade) and offer /trader-budget to set the user's own. Never treat example sizes as the user's.
2. Run what they asked: run TICKER, or scan. Before 9:50 AM ET, after 3:30 PM ET, or on a closed day there is no live setup: offer a replay of the latest session at 10:15 with --at-trigger.
3. Tell the user, from the printed text: the status line per ticker, the time machine's odds and verdict line exactly as printed, the size and the risk verdict, and on a replay what really happened. Say the seed.
4. Open the dashboard (warroom.py open) and say where it is.
5. If one passed live and they want it: it is already promoted (run) or pick TICKER (scan). Hand off to /trader-broker. Offer /trader-alerts add --from-plan for the stop, the target and the flat-by time.

Rules

• Every number comes from the pack's tools. Repeat their verdict lines, do not rewrite them. Never round a coin flip up into an edge.
• Say "matched a rule" and "ranked by the replays", never "a buy", "a winner" or "the best trade".
• Never print a P&L, a return or a profit figure. R and the max loss at the stop are plain plan numbers.
• A replay is never an order. A board from another day is never promoted.
• Simulated odds are not predictions. Say it once, plainly.

What it never does

It never places an order, never calls a broker tool, never changes a limit, never edits an order to make it pass, and never promotes a replay or a played out plan.

Safety

• Education and tooling, not financial advice. A PASS is a rules check, not a recommendation.
• Paper trading by default ("paper": true). Going live is the user's own decision, in their own config.
• Nothing in this pack places an order except trader-broker, and only when all three hold: trader-risk passed that exact order today, the user typed an explicit yes for that one order in the chat, and the user clicked Allow once on Claude's own card because the order tool is set to "Needs approval" (Customize > Connectors > tool permissions). Never set an order tool to "Always allow".
• No options, margin, short selling or leverage unless the config allows it and the user confirms.
• Broker keys live in environment variables or the Claude Desktop config, never in the chat.