Card games & deckbuilders
Play card games and deckbuilders to a goal.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Slay the Spire · solitaire
- Class GUI-need label
G2· GUI-essential
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R13.02.01.02-T01 Reach floor 10 in Slay the Spire on a fixed seed
Original instruction
Slay the Spire (desktop) is at the main menu. Start a standard run as the Ironclad at Ascension 0 with seed 'AGENTRUN1' and reach floor 10 alive, choosing card rewards, paths and events yourself.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Slay the Spire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Run autosave/run history file parsed: seed AGENTRUN1, character Ironclad, floor ≥ 10 with HP > 0. Mods (ModTheSpire, CommunicationMod) are not installed; the agent has no shell, and writes to the save folder are logged by process, so a save not written by the game is detected.
Proposed partial credit
50% floors 1-5 reached (a tenth each); 50% floors 6-10 reached (a tenth each)
R13.02.01.02-T02 Win a seeded game of Klondike solitaire by dragging cards
Original instruction
AisleRiot (GNOME Solitaire) is open with Klondike, draw one, game seed 1234 (a winnable deal). Move all 52 cards to the foundations by dragging cards and stacks; undo is allowed.
- Platforms
lin· Linux- Applications
- solitaire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropvis· Visual judgmentclick· Point & click- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
AisleRiot game-state file and statistics: game seed 1234 marked won. Keyboard-driven 'auto move' is allowed only for final foundation moves; no external solver input (no clipboard or scripting interface exists).
Proposed partial credit
50% scaled by the first 26 cards on foundations; 50% scaled by the remaining 26 (52 = full)
R13.02.01.02-T03 Solve FreeCell deal number 1
Original instruction
Microsoft Solitaire Collection (Windows) or KPatience (macOS/Linux build) is open on FreeCell deal #1. Solve the deal by moving all cards to the foundations.
- Platforms
win· Windowsmac· macOS- Applications
- solitaire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
App statistics / saved game: deal #1 recorded as won during the session; screen OCR of the win dialog as fallback. The agent has no shell, and the harness logs which process writes the stats/save file, so a win not recorded by the app is detected.
Proposed partial credit
50% scaled by the first 26 cards on foundations; 50% scaled by the remaining 26
R13.02.01.02-T04 Remove a Strike at the Slay the Spire shop
Original instruction
Slay the Spire is loaded from a mid-run save where the player stands in a shop with 200 gold. Use the shop's card-removal service to remove one 'Strike' and leave the shop without buying anything else.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Slay the Spire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 1 / 5
Proposed verifier
Autosave after leaving the shop: deck has one fewer Strike, gold = 200 − removal cost, no other deck/relic/potion changes. No mods are installed; the agent has no shell, and writes to the save folder are logged by process, so a save not written by the game is detected.
Proposed partial credit
60% Strike removed; 40% no other purchases
R13.02.01.02-T05 Defeat the Act 1 boss in Slay the Spire
Original instruction
Slay the Spire, Silent, Ascension 0, seed 'AGENTBOSS1'. Play through Act 1 and defeat its boss.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Slay the Spire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Run history: floor 17 boss killed (act 2 entered) for this seed and character. No mods installed; the agent has no shell, and writes to the save folder are logged by process, so a run history not written by the game is detected.
Proposed partial credit
70% scaled by floor reached up to 16; 30% boss defeated
R13.02.01.02-T06 Beat Ante 4 in Balatro on White Stake
Original instruction
Balatro (desktop), new run with the Red Deck, White Stake, seed 'AGENT4'. Select and play poker hands, buy jokers, and beat the Ante 4 boss blind.
- Platforms
mac· macOSwin· Windows- Applications
- Balatro
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Balatro save (profile and run .jkr, decompressed) parsed: seed AGENT4, ante ≥ 5 reached or Ante 4 boss beaten. Lovely/Steamodded loaders absent; the agent has no shell, and writes to the save folder are logged by process, so a .jkr not written by the game is detected.
Proposed partial credit
50% scaled by blinds 1-6 beaten; 50% scaled by blinds 7-12 beaten, ending with the Ante 4 boss
R13.02.01.02-T07 Win a two-suit game of Spider Solitaire
Original instruction
Microsoft Solitaire Collection on Windows (offline, test profile) or a Spider build on macOS: start Spider, Medium (2 suits), seeded deal. Complete all 8 sequences.
- Platforms
win· Windowsmac· macOS- Applications
- solitaire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropvis· Visual judgmentclick· Point & click- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Game statistics/save show the seeded deal won. Fallback: win screen OCR. The agent has no shell, and the harness logs which process writes the stats/save file, so a win not recorded by the app is detected.
Proposed partial credit
50% scaled by sequences 1-4 completed; 50% scaled by sequences 5-8 completed
R13.02.01.02-T08 Build a legal deck from a pool and win a best-of-three in Forge
Original instruction
Forge (Magic: The Gathering engine, desktop) opens a Sealed event with the card pool in input/pool.dck. Build a 40-card deck using the deck editor (lands allowed), then win a best-of-three match against the AI opponent.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Forge
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Saved deck file: 40+ cards all from the pool (plus basic lands). Forge match log: 2 game wins before 2 losses. Forge's dev mode is disabled in preferences.
Proposed partial credit
30% legal deck from pool; 35% first game won; 35% second game won
R13.02.01.02-T09 Finish top two in a play-money PokerTH sit-and-go
Original instruction
PokerTH (desktop) local game: 6 players (you + 5 computer opponents), play-money chips 5,000 each, blinds rising every 10 hands. Finish in 1st or 2nd place.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- PokerTH
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
PokerTH log database (local SQLite) parsed: final rank of the human seat ≤ 2. Play money only; no network game. The agent has no shell or sqlite3, and the log database's writing process is recorded, so a rank not written by PokerTH is detected.
Proposed partial credit
50% scaled by finishing place from 6th to 4th; 50% scaled from 4th to 2nd (1st also full)
R13.02.01.02-T10 Win a game of Hearts against three computer players
Original instruction
A Hearts game (Windows: Microsoft Hearts-style app; macOS/Linux: a local Hearts app) is open with 3 AI opponents, game to 100 points. Pass cards, follow suit and win (lowest score when someone reaches 100).
- Platforms
win· Windowsmac· macOS- Applications
- Hearts
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
App's game log/score file: player has the lowest total when the game ends. Fallback: final score screen OCR. The agent has no shell, and the harness logs which process writes the score file, so a result not recorded by the app is detected.
Proposed partial credit
30% game completed; 70% win
R13.02.01.02-T11 Win a Covenant 0 run in Monster Train
Original instruction
Monster Train, clans Hellhorned + Awoken, Covenant 0, fixed seed. Defend the Pyre through all 8 rings and defeat the final boss (Seraph).
- Platforms
win· Windowsmac· macOS- Applications
- Monster Train
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Save/run history: run victory flag with the seed. No mods loaded; the agent has no shell, and writes to the save folder are logged by process, so a run history not written by the game is detected.
Proposed partial credit
80% scaled by rings 1-7 cleared; 20% Seraph defeated in the final ring
R13.02.01.02-T12 Survive the first map of Inscryption's Act 1
Original instruction
Inscryption, new game from the cabin start. Play card battles against Leshy through the first map region and defeat its boss (the Prospector).
- Platforms
win· Windowsmac· macOSlin· Linux- Applications
- Inscryption
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Save file parsed: Act 1 map region index advanced past the Prospector. No mods; the agent has no shell, and writes to the save folder are logged by process, so a save not written by the game is detected.
Proposed partial credit
60% scaled by battles won on the map; 40% Prospector defeated
R13.02.01.02-T13 Reach Act 2 in Slay the Spire with a deck of at most 15 cards
Original instruction
Slay the Spire, Defect, Ascension 0, seed 'THINDECK'. Reach Act 2 (floor 18) with a deck of 15 or fewer cards when entering it, by skipping rewards and removing cards.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Slay the Spire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Run autosave at floor 18: character Defect, seed THINDECK, deck size ≤ 15. No mods are installed; the agent has no shell, and writes to the save folder are logged by process, so a save not written by the game is detected.
Proposed partial credit
60% scaled by floor reached up to 18; 40% deck ≤ 15 at Act 2
R13.02.01.02-T14 Clear a Pyramid solitaire board
Original instruction
Microsoft Solitaire Collection (Windows, offline test profile) or a Pyramid solitaire app on macOS is open on a seeded Pyramid deal. Remove all cards from the pyramid by pairing cards that sum to 13.
- Platforms
win· Windowsmac· macOS- Applications
- solitaire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
App statistics/save: seeded deal won; fallback: win dialog OCR. The agent has no shell, and the harness logs which process writes the stats/save file, so a win not recorded by the app is detected.
Proposed partial credit
50% scaled by the first 14 pyramid cards removed; 50% scaled by the remaining 14
R13.02.01.02-T15 Score 10,000 chips in one Balatro hand by Ante 3
Original instruction
Balatro, Red Deck, White Stake, seed 'BIGHAND'. By the end of Ante 3, play a single hand that scores at least 10,000 chips (score shown after the hand).
- Platforms
mac· macOSwin· Windows- Applications
- Balatro
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Profile stats file: best hand score ≥ 10,000 recorded during a run with seed BIGHAND and ante ≤ 3 (run snapshot). Mod loaders are absent; the agent has no shell, and writes to the save folder are logged by process, so stats not written by the game are detected.
Proposed partial credit
50% scaled by best hand from 1,000 to 5,500 chips; 50% scaled from 5,500 to 10,000
R13.02.01.02-T16 Make three of five contracts as declarer in a bridge program
Original instruction
Bridge Baron (desktop, robot opponents) deals 5 preset hands where you are declarer in a stated contract. Play each hand, choosing cards for declarer and dummy, and make at least 3 contracts.
- Platforms
win· Windowsmac· macOS- Applications
- Bridge Baron
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Exported PBN with play records: tricks taken per hand vs contract level; double-dummy solver confirms the hands were makeable. The agent has no shell or text editor, and the PBN's writing process is logged, so a file not exported by Bridge Baron is detected.
Proposed partial credit
20% first contract made; 20% second; 20% third; 40% fourth and fifth
R13.02.01.02-T17 Win a game of Gin Rummy to 100 against the computer
Original instruction
A Gin Rummy app (local, play money) is open against the computer. Draw, discard and knock/gin to win a game to 100 points.
- Platforms
win· Windowsmac· macOSios· iOS/iPadOS- Applications
- Gin Rummy
- Interfaces
desk· Desktop GUImob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
App stats file or final score screen OCR: player reaches 100 first. No shell is available on any device, and on desktop the stats file's writing process is logged, so a win not recorded by the app is detected; iOS relies on the score-screen OCR.
Proposed partial credit
30% scaled by hands won; 70% game won
R13.02.01.02-T18 Clear a Mahjong solitaire layout
Original instruction
KMahjongg (desktop) is open on the 'Default' layout with a solvable seeded deal. Remove all tiles by matching free pairs. Hints are disabled; undo is allowed.
- Platforms
lin· Linuxwin· Windowsmac· macOS- Applications
- Mahjong solitaire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
KMahjongg saved game and highscore table: game number solved with 0 tiles left. The agent has no shell, and writes to the KMahjongg config folder are logged by process, so a result not written by the app is detected.
Proposed partial credit
50% scaled by the first 72 tiles removed; 50% scaled by the remaining 72 (144 = full)
R13.02.01.02-T19 Defeat The Guardian while taking at most 20 damage
Original instruction
Slay the Spire, Watcher, loaded from a save at the Act 1 boss door (The Guardian) with a prepared deck. Defeat the Guardian taking 20 or less total damage in the fight, managing stances and the Guardian's defensive mode.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Slay the Spire
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Autosave after the fight: boss defeated and HP loss = HP before − HP after ≤ 20. No mods are installed; the agent has no shell, and writes to the save folder are logged by process, so a save not written by the game is detected.
Proposed partial credit
60% boss defeated; 40% scaled by damage taken (40 → 20)
R13.02.01.02-T20 Win a game of Uno against three bots on a phone
Original instruction
On the iOS Simulator, a local Uno-style card app is open against 3 bots, standard rules. Play cards with taps and drags, call 'Uno' with one card left, and win the game.
- Platforms
ios· iOS/iPadOS- Applications
- Uno-style app
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesvis· Visual judgment- Work product
WP11· Game/sim state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
App's local game log (SQLite in the app container): player winner; no missed-Uno penalty. Accessibility-driven automation of the app is disabled.
Proposed partial credit
30% game completed; 20% Uno called correctly; 50% win
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.