Modegate: your agent reads intent before it acts

Stop wasting tokens and time on scope drift, premature edits and excessive politeness. This skill implements a strict collaboration discipline that forces the model to listen the first time, preventing the "while I was there" refactoring trap and eliminating unnecessary conversational padding. It introduces a rigorous execution framework that prioritizes outcomes over explanations.

What it does

• Enforces Outcome-First Communication mandates starting with results and constraints while stripping out rationale unless requested.
• Identifies Decision Modes categorizes every prompt into Discuss, Build, Pause, or Pivot to prevent premature execution.
• Recognizes Frustration Signals triggers an immediate halt and hard pivot when the user provides negative feedback, stopping useless iterations.
• Deploys Pre-Action Gates runs a four-question risk assessment before performing any non-trivial or destructive operation.
• Eliminates Conversational Tax removes preambles, "Great question!" fillers, and repetitive summaries to focus on code and data.

How it works

1. Context Initialization the agent scans project manifests and configs to adopt your specific technical voice and language.
2. Mode Detection the agent classifies your input (e.g., "propose plan" vs "ship it") to determine if it should write code or just talk.
3. Risk Gating before running commands like rm or git reset, the agent internalizes a cost-of-error check.
4. Direct Execution deliverables are provided without introductory fluff or closing summaries, ending only with an action-verb next step.

Frameworks & tools

This skill is platform-agnostic and works across any IDE or CLI environment including Claude Code, Cursor, and GitHub Copilot. It is designed to compose with sister tools like Tokenlean for token compression.

Why this beats prompting it yourself

Manual prompts often fail to stop an LLM from "helping" too much by refactoring adjacent files or apologizing for errors. This skill encodes these boundaries as a persistent discipline, ensuring the agent pivots instead of arguing when you give negative feedback.

Use cases

• Rapid Prototyping when you need to build exactly what is spec'd without tangential code changes.
• Risk-Sensitive Refactoring ensuring destructive CLI operations are always preceded by an explicit authorization check.
• Token-Efficient Deep Work cutting the conversational fluff that inflates context and cost.

Known limitations

Works best with a little project context (README, manifest). It does not auto-route to sister skills; those must be triggered by their own specific keywords.