AI Productivity Kit | AI Productivity

cursor — ai-productivity

user@cursor ~ %

How it works

Session Kickoff

Paste once. Sets SPEC-first protocol, 85% gate, and one-question rule.

Context Pack

3–7 files and switches. Grounds the AI in your repo, no guessing.

Router

Routes to Plan, Agent, Debug, or Ask. Modes never mix.

Plan → Agent

Plan first. Code second. Security checks when it matters. Ship.

confidence gate before any action

question asked at a time, then it stops

invokable Skills for every phase of the workflow

agent lenses covering the full stack

PROTOCOL

The 85% confidence gate means the AI either acts or asks — never guesses. If unsure, it asks exactly one question and waits.

MODES

Plan, Agent, Debug, and Ask — Cursor-native modes that map directly to the kit. Planning and coding never mix.

GROUNDING

Clarify the problem and acceptance criteria before code. The AI reads your docs/ai/ai-config.md. No invented paths or APIs.

ADVISORIES

Every response starts with Mode: Plan | Agent | Debug | Ask, recommended model, context risk, and switch recommendations so you stay in control.

SECURITY

For high-risk work like auth, uploads, or exports, the AI asks one security question and stops until you answer.

PORTABLE

Cursor rules, GitHub Copilot instructions, and Google Antigravity rules — all pre-wired with 8 invokable Skills and shipping in the same zip.

8 invokable Skills, security enforcement hooks, MCP trust posture, indirect-injection defence, and Google Antigravity support.

Added dark mode support, refined spacing, and better mobile navigation across all doc pages.

New security checklists and a hard stop gate for sensitive actions like uploads or auth changes.

Copy the kit into your repo. Open Cursor. Paste the Session Kickoff.