graph-based plan-building IDE with cognitive operations.
canvas is not a claude wrapper. it's a new interface primitive: reasoning-as-graph.
linear chat forces you to think linearly. canvas lets you branch, stress-test, and refine plans as a visual structure — then execute them.
plans are the highest-leverage artifact in agentic coding. a good plan = claude executes well. a bad plan = you watch claude flail for 10 minutes before ctrl-c.
but plan-building in linear text is brutal:
- you can't see the structure
- you can't click into a substep and stress-test it
- you can't branch "what if we did X instead of Y" without losing the original
- you can't cross-link dependencies between steps
runeforge canvas fixes this.
- paste rough goal into canvas as root
- click
@excavate→ assumptions surface as children - click an assumption →
@antithesize→ opposition spawns - click best path →
@stressify→ failure modes surface - iterate until plan feels robust
- press
x→ execute plan in claude code - press
r→ review execution, surface divergence
| operation | what it does |
|---|---|
| @excavate | surface hidden assumptions |
| @antithesize | generate standalone opposition |
| @synthesize | compress conflicting positions |
| @stressify | probe for failure modes |
| @simulate | trace execution forward |
| @diverge | generate alternative approaches |
| @dimensionalize | map to measurable dimensions |
| @negspace | detect what's conspicuously absent |
| @metaphorize | map from familiar domain |
| @rhyme | fast structural similarity |
39 skills in extended set. request access →
pip install runeforge-canvasor from source:
git clone https://github.com/jordanrubin/ft-ui
cd ft-ui
pip install -e .runeforge-canvas # new canvas
runeforge-canvas my-plan.json # load existing| key | action |
|---|---|
q |
quit |
s |
save |
e |
export plan as markdown |
x |
execute plan in claude code |
r |
review execution |
n |
new canvas |
esc |
focus operations |
- python 3.11+
- claude code installed and authenticated (
claudecommand available)
MIT
built for future tokens