genius

genius

Injects a structured engineering identity into Claude Code and Cursor — same reasoning instincts, every session, without re-prompting.

Why

Without identity: Claude reads what's in front of it and executes the path of least resistance. With identity: it observes before acting, validates assumptions, and flags irreversible moves first.

Tested: same model, same repo, same one-line prompt, bypass permissions, both unattended.

	no identity	genius
Time	~13 min	~34 min
Cost	~$8	~$13
Lines of code	more	~100 fewer
Behavior	read repo, executed a stale CLAUDE.md instruction	deliberated, stayed in step 2 while no-identity was on step 5, ran independent validation

Backed by research: semi-formal structured reasoning outperforms standard prompting by 8–11pp on code tasks (arxiv 2603.01896).

Quickstart

Via plugin (recommended):

/plugin marketplace add shaquan1228/genius
/plugin install genius@genius
/reload-plugins

Manually:

git clone [email protected]:shaquan1228/genius.git
cd genius
bin/setup

For Cursor: paste the printed snippet into Cursor > Settings > Rules for AI.

How it works

Once installed, a SessionStart hook automatically injects docs/identity.md into every Claude Code session — no commands needed. The identity defines how the model should observe, decide, and act. Cursor gets the same context via .cursor/rules/.

docs/identity.md is the only file you need to understand or customize.

Run /genius:think at any point to have Claude explicitly load and summarize the reasoning framework.

Commands

Command	Description
/genius:think	Load and summarize the genius reasoning framework
setup	Manually inject identity into Claude Code and Cursor
teardown [--confirm]	Remove the injection (dry-run by default)