fablize-for-cursor is a port of fivetaku/fablize (a Claude Code plugin) to Cursor.
Same procedures: run and observe what you built, investigate with competing hypotheses, finish multi-step work with proof, don’t stop on a promise.
Wiring is Cursor-native. Project rules inject discipline, project hooks enforce.
Install: Install: clone, open as a trusted workspace. .cursor/rules/ and .cursor/hooks.json load on their own. To gate another repo, copy .cursor/hooks.json, .cursor/hooks/, .cursor/rules/, scripts/gate/, and scripts/goals.py into that project root. Don’t also install it as a plugin in the same workspace or the gate runs twice.
Hooks record edits and shell results, follow up when a deep change has no observed verification, catch “I’ll do it” without doing it, and nag when the same failure class repeats.
Honest limits: this does not raise model capability. Effect numbers from the original Opus-vs-Fable measurement do not apply to whatever model you pick in Cursor. Rule routing is agent-selected from each rule’s description, not a deterministic pack inject every prompt. beforeSubmitPrompt only classifies and resets the ledger in current beta. Cloud agents load project hooks, not ~/.cursor/hooks.json.
Marketplace listing is submitted; until Customize shows fablize-for-cursor, use the clone path above. Issues and PRs welcome!