⚙️ @infandev/agent-kit

🚀 Installation

Install and compile the AI context engine directly into your workspace with a single command. You must specify your IDE target.

For Cursor Developers

Generates native .cursor/rules/*.mdc files and securely connects Master Context modules (skills & workflows).

npx @infandev/agent-kit init --cursor

For Windsurf Developers

Compiles the AGENTS.md unified persona matrix and initializes .windsurfrules.

npx @infandev/agent-kit init --windsurf

For VS Code Developers (Copilot / Cline / RooCode)

Generates a unified .github/copilot-instructions.md configuration natively enforcing Tier 0 globals alongside your architecture.

npx @infandev/agent-kit init --vscode

For Antigravity

Initializes the .agents folder containing the pure source context for tools like Aider or RooCode.

npx @infandev/agent-kit init --antigravity

🆚 Why Infandev Agent Kit? (Vanilla AI vs Enterprise AI)

| Feature | Vanilla LLM / Default IDE | Infandev Kit (v2.0) | | :--- | :--- | :--- | | Hallucination Risk | High (Blind assumptions) | Zero (TNbN + Grounding Gates) | | Context Window | Overloaded (All rules always on) | Efficient (Progressive Disclosure) | | Strategic Alignment | Drifts easily | Locked in (Top-down Intent System) | | Reasoning Depth | Static (Tries to over-architect typos) | Dynamic (Scales based on Effort) | | Confidence | Confidently wrong | Honest (Explicit Uncertainty Protocol) |

⚡ The Architecture (v2.0 Enterprise Edition)

Infandev Kit acts as the intelligence layer for AI-native development by standardizing agentic reasoning across platforms. With the v2.0 Enterprise upgrade, the kit now scales reasoning depth, guarantees strategic alignment, and strictly prevents hallucinations.

🎯 Intent System (New in v2.0)

The kit now implements top-down Intent Engineering. A global project-intent.md defines your project's non-negotiable trade-offs, which are dynamically loaded by all 20 specialized agent profiles. When you pivot strategy, the entire AI workforce adapts instantly.

⚖️ Effort Calibration (New in v2.0)

Not every task requires deep architectural planning. The kit now maps the effort (low/medium/high) of your request to the required analytical depth. Typo fixes get immediate execution; system architecture gets multi-pass verification.

🧠 TNbN Protocol + Grounding Gates (Zero Hallucination)

This package forces the LLM to follow a strict Node-based operational pipeline, saving thousands of tokens. In v2.0, this is reinforced by new Grounding Gates:

1. 🌳 State Node: Declare the target and assumptions before calculating.
2. 🔍 Verify Node: Use raw filesystem tools to confirm the actual landscape.
3. 🛡️ Grounding Gate: Quote verbatim code lines before making analytical claims. Express explicit Uncertainty (❓, 🔶) instead of guessing.
4. 🛠️ Act Node: Perform the exact atomic modification required.
5. ✅ Confirm Node: Verify the execution with a final tool validation, respecting Context Freshness tags to prevent drift.

📦 Progressive Disclosure

Unlike default IDE setups that overwhelm the AI with rule vomit, @infandev/agent-kit dynamically connects targeted Agent glob triggers to deep Skill modules (@[.cursor/skills/...]). As your AI reads a database schema, it actively pulls the database-architect context.

🛠️ Configuration Guide (v2.0 Features)

1. Setting Your Project Intent

After initializing the kit, define your project's non-negotiable trade-offs to align your entire AI workforce.

1. Open .agents/.shared/project-intent.md
2. Modify the Primary Mandate and Trade-off Resolution sections.
3. Every specialized agent (Frontend, Backend, etc.) will automatically read and adhere to this strategy on every invocation.

2. Effort Calibration

The kit automatically maps your requests to the optimal reasoning depth.

• Low Effort: (e.g. Fix typo) → Immediate execution. No planning overhead.
• Medium Effort: (e.g. Refactor component) → Standard TNbN + Code Quoting.
• High Effort: (e.g. Build new feature) → Full Grounding Gates, multi-pass verification.

MIT © 2026 Infantex Team