gtow-trainer-drill

name

gtow-trainer-drill

description

Navigate the GTO Wizard SPA over a live CDP session and build precise GTOW Trainer drill deep-links. Use when you need to (a) reverse-engineer what URL params a GTOW page uses, (b) drive the Practice/Trainer UI from the terminal, or (c) construct a `/practice/trainer?...` deep-link that lands the player on an exact spot (position + action + IP/OOP + street). Triggers: "gtow trainer 參數", "drill 連結", "how does the trainer URL encode X", "reverse engineer gtow drill", "build a precise practice link".

GTO Wizard Trainer drill deep-links (CDP navigation)

The companion to gtow-cdp-session (which handles attaching). This skill is the hard-won map of how the SPA behaves and the Trainer drill URL contract, so you never have to trial-and-error the UI again.

0. Attach first

Reuse the dedicated debug profile from gtow-cdp-session. A CDP Chrome usually persists on port 9222 already:

curl -s -o /dev/null -w "%{http_code}\n" http://127.0.0.1:9222/json/version   # 200 == up
agent-browser --cdp 9222 tab list          # authed when NOT on /login

If it's down or logged out, follow gtow-cdp-session to relaunch + log in.

1. The Trainer drill URL contract (VERIFIED 2026-07-09)

When you configure a drill and click START TRAINING, the SPA navigates to https://app.gtowizard.com/practice/trainer?.... Two classes of params:

Config params (the drill filter — this is what you build to pin a spot):

Param	Meaning	Values
fh_hero	hero position(s)	UTG,UTG+1,LJ,HJ,CO,BTN,SB,BB — comma = "any of"
fh_opponent	opponent position(s)	same set, comma-separated
fh_rel_positions	relative position	IP / OOP
fh_actions	preflop action(s)	RFI, vsSRP (= "vs Open"), vs3bet, vs4bet, vsRaiseCall, vsSqueeze, vsLimp, vsIso, StartOfHand (= "From start") — comma-separated
fh_start_spot	where the drill starts	preflop / flop / custom_spot
preflop_actions	exact preflop line (custom_spot)	e.g. F-F-F-R2-F-F-F-C
flop_actions	exact flop line (custom_spot)	e.g. X-R1.8-R4.8
depth / depth_list	stack depth (bb+0.125)	20.125; list is CSV of allowed depths
gametype	solution set	MTTGeneral (MTT ChipEV), MTTMRonlyGeneral, ...
fh_groups / fh_groups_selection	pinned hero hand range	manual + CSV of AKs,22,...

Instance params (the dealt hand, NOT filter — ignore when building links): fh_hero_tmp, fh_opponent_tmp, fh_rel_position_tmp, fh_action_tmp, hand_tmp, board_tmp, history_spot, and the realized preflop_actions / flop_actions for that one hand. The _tmp suffix always = "this hand".

Label → value gotchas: the UI label "vs Open" encodes as fh_actions=vsSRP; "From start" = StartOfHand. vsIso/vsLimp/vs3bet/vs4bet/vsRaiseCall/ vsSqueeze/RFI match their labels. A flop drill with a specific action line uses fh_start_spot=custom_spot + flop_actions=..., not fh_start_spot=flop.

Baseline UI flags (ranges visible, etc.) live in scripts/gtow_trainer_url.py _TRAINER_UI_DEFAULTS. That builder currently pins only fh_actions + depth + fh_start_spot ("Option Y"); the params above are the deferred "Option Z" — extend it with fh_hero/fh_opponent/fh_rel_positions to pin the exact spot.

2. Position category → GTOW positions

Repo taxonomy (scripts/spot_taxonomy.py) → fh_hero/fh_opponent value: EP=UTG,UTG+1 · MP=LJ,HJ · LP=CO,BTN · SB=SB · BB=BB. For an exact-hero line (RFI/vsOpen) pass the single position; for a category line pass the CSV.

3. How to drive the SPA (what actually works)

The Practice UI is a React SPA. Lessons that save loops:

• Click by text via eval, not by coordinates. Icon buttons are SVGs; document.elementFromPoint(x,y).click() hits a wrapper div and does nothing. Match visible text instead:

  agent-browser --cdp 9222 eval '(()=>{const e=[...document.querySelectorAll("*")]
    .find(x=>x.textContent.trim()==="START TRAINING"&&x.offsetWidth>0&&x.children.length<=2);
    if(e){e.click();return"ok"}return"miss"})()'
  

  Constrain with children.length<=1/2 so you get the leaf element, not an ancestor wrapping the whole page.
• The "New" button's label is an 8px <span>; click its BUTTON ancestor (walk up ~4 parents to tagName==="BUTTON").
• Reaching the drill config page: from /practice/drills → click New → ensure Hold'em (not PLO) → the config page (visual table + Starting spot + Preflop action + ALL SETTINGS + START TRAINING) appears. Clicking a drill name in the left sidebar loads that saved drill's full pinned config — the fastest way to see how a precise spot is encoded. ("Configure Training" text match is flaky; clicking a sidebar drill name works reliably.)
• Read the resulting URL after START with agent-browser --cdp 9222 get url, then grep -oE 'fh_[a-z_]*=[^&]*' to isolate the config params.
• open <url> forces fresh state; history.back() on the SPA often lands on the running trainer, not the config page — prefer open + re-drive.
• Screenshot to orient (agent-browser --cdp 9222 screenshot /tmp/x.png then Read it) whenever a text-match misses — the header Hold'em/PLO toggle and the current route are easy to lose track of.
• Saved drills that encode useful configs to inspect: "Mastering the BTN" (fh_hero=BTN, fh_actions=vsSRP,RFI), "IP Flop C-Bet raised" (fh_rel_positions=IP, fh_start_spot=custom_spot, flop_actions=X-R1.8-R4.8), "Blind vs Blind" (fh_hero=SB,BB, fh_opponent=BB,SB).

4. Example — build a precise deep-link

Spot: hero BB defending OOP vs a late-position open, ~20bb, preflop.

https://app.gtowizard.com/practice/trainer?solution_type=gwiz&gametype=MTTGeneral
  &depth=20.125&depth_list=20.125&fh_start_spot=preflop&fh_actions=vsSRP
  &fh_hero=BB&fh_opponent=CO,BTN&fh_rel_positions=OOP
  &fh_groups_selection=manual&...(_TRAINER_UI_DEFAULTS)...&dialogs=

Opening it drops the player straight into that drill; GTOW deals hands within the filter (each dealt hand adds *_tmp params).

5. Teardown

Leave Chrome running for the next session (profile + login persist). To stop: pkill -f "user-data-dir=$HOME/.config/gtow-cdp".