Showcase / Made at Kane CLI Hack Day
Other
Room for Error
AI-generated escape rooms that Kane CLI test-plays before players can finish them.
Room for Error turns one theme into a short illustrated point-and-click escape room with rooms, items, puzzles, clues, and two endings. An OpenAI director writes the scenario, FLUX paints each scene, and a vision model locates interactive objects. Kane CLI plays every escape route in an isolated browser session. A con…
2 builders credited · submitted Oct 24, 6:22 PM PDT
Kane · proof view
- Generate and validate scenario data
- Run Kane CLI on the fixed route objective
- Classify pass, game failure, off-route, or infrastructure error
- Repair scenario data and rerun, or release the passed route
- Go to http://localhost:5180/preview/018aa10e-a127-40ed-8732-8767a809355b. This is an escape-room game. Act only with the labelled buttons listed under 'Things to explore' below the picture; ignore the picture itself. Read the narration text after every action: it contains clues. To enter a code, type it into the keypad's text field exactly as the clue gives it and press 'Submit code'. Your goal is the 'Rooftop Escape' ending. The room has other exits: never use any exit or ending other than theSuspected · failed
- Go to http://localhost:5180/preview/44cb36e1-4738-4e6b-a950-50ed1f38304f. This is an escape-room game. Act only with the labelled buttons listed under 'Things to explore' below the picture; ignore the picture itself. Read the narration text after every action: it contains clues. To enter a code, type it into the keypad's text field exactly as the clue gives it and press 'Submit code'. Your goal is the 'Rooftop Escape' ending. The room has other exits: never use any exit or ending other than theSuspected · errored
- Go to http://localhost:5180/preview/65aa850a-8dab-44f6-bb1e-dc87a09eed6d. This is an escape-room game. Act only with the labelled buttons listed under 'Things to explore' below the picture; ignore the picture itself. Read the narration text after every action: it contains clues. To enter a code, type it into the keypad's text field exactly as the clue gives it and press 'Submit code'. Your goal is the 'Tunnel Escape' ending. The room has other exits: never use any exit or ending other than the oSuspected · errored
Product
Other
Run command in repo
Harness
4 Kane parts
React + Node game engine · OpenAI Chat Completions · gpt-6.1-sol · FLUX scenes + vision hotspots
Evidence
3 runs
0 verified · 3 detected
Repo
c5cec8a
github.com/Anmol-tech/room-for-error
01 / The creation
What it is, and why.
AI-generated games can be structurally valid while a clue, lock, item, or route still fails during real interaction. In the museum mission, the validator found zero issues even though the desk note showed 4471 and the archive terminal expected 4417, so the player could not finish.
Room for Error turns one theme into a short illustrated point-and-click escape room with rooms, items, puzzles, clues, and two endings. An OpenAI director writes the scenario, FLUX paints each scene, and a vision model locates interactive objects. Kane CLI plays every escape route in an isolated browser session. A confirmed game failure keeps the route closed, sends evidence to the director for a data-only repair, and reruns the same objective on the new version.
- 01Enter a one-line theme for the escape room.
- 02The OpenAI director writes scenario JSON and the engine renders a playable room.
- 03Kane CLI plays each ending in an isolated browser session tied to that version.
- 04A failed route returns for data-only repair; a passed route opens to players.
02 / The harness
How Kane checks it.
A creator enters a theme. The OpenAI director writes scenario data for the game engine. Kane CLI follows the visible clues and controls in an isolated preview. Its version-bound result decides whether each route opens to players or returns to the director for repair.
- React + Node game engine
- OpenAI Chat Completions · gpt-6.1-sol
- FLUX scenes + vision hotspots
- Kane CLI isolated route playthroughs
- 01Generate and validate scenario data
- 02Run Kane CLI on the fixed route objective
- 03Classify pass, game failure, off-route, or infrastructure error
- 04Repair scenario data and rerun, or release the passed route
03 / Evidence
What actually ran.
Verified means a passing run reported through a paired, verified channel. A failing run is a detection, not a confirmed bug. Everything else is as the team reported it.
- Suspected · failedOct 24, 2:44 PM PDT
Go to http://localhost:5180/preview/018aa10e-a127-40ed-8732-8767a809355b. This is an escape-room game. Act only with the labelled buttons listed under 'Things to explore' below the picture; ignore the picture itself. Read the narration text after every action: it contains clues. To enter a code, type it into the keypad's text field exactly as the clue gives it and press 'Submit code'. Your goal is the 'Rooftop Escape' ending. The room has other exits: never use any exit or ending other than the
kanehack_cli · kanehack-047d386b-a607-4eae-86d9-943a4ddf94a7 · exit 1 · 405c946
- Suspected · erroredOct 24, 2:49 PM PDT
Go to http://localhost:5180/preview/44cb36e1-4738-4e6b-a950-50ed1f38304f. This is an escape-room game. Act only with the labelled buttons listed under 'Things to explore' below the picture; ignore the picture itself. Read the narration text after every action: it contains clues. To enter a code, type it into the keypad's text field exactly as the clue gives it and press 'Submit code'. Your goal is the 'Rooftop Escape' ending. The room has other exits: never use any exit or ending other than the
kanehack_cli · kanehack-383ab642-92af-4f9b-befb-6ee0cd29ce04 · exit 0 · 405c946
04 / Outcome
What changed.
- Qualitative
A route reaches players only after Kane passes that exact scenario version.
Source: Version-bound gating in the submitted app and README (team reported)
- Qualitative
The museum code mismatch was repaired in scenario data while both route objectives stayed fixed.
Source: Repair record from museum v4 to v5 and the bound Kane claims (team reported)
- Qualitative
Infrastructure errors and off-route runs remain separate from confirmed game failures.
Source: Run classification and repair guards in the submitted app (team reported)
Known limitations
- The app is local-only; new adventure generation requires OpenAI and BFL API keys.
- Kane runs are serialized because parallel runs were observed sharing a browser session.
- The Studio currently labels the bound run reports as team-reported rather than source-verified.
05 / Ship it forward
The code. The proof.
The story.
- Anmol Sharma
- Siddharth Bhat
Also submitted
Run it
npm install && cp .env.example .env && npm run dev