Token Master for AI Coding

The problem

When an agent handles refactoring, debugging, or code review, the cost issue is often not missing capability but redundant context and output: reading irrelevant directories, keeping stale conversation, emitting whole files for small edits, repeating continuations, and adding unnecessary abstractions. This skill provides a token-efficient development protocol for AI-assisted coding tasks, aiming to reduce wasted tokens without changing functionality or quality.

How it works

It enforces four categories of constraints:
- Input throttling: load only task-relevant functions, snippets, and error logs; filter unrelated comments and stale history.
- Output throttling: prefer file path + line numbers + diff; do not resend earlier content during continuation; implement only the minimum viable change.
- Process control: start complex tasks with a concise step outline; cap parallel subagents at 3; run one key validation pass after output.
- Quality gate: self-check before responding to avoid whole-file output, repeated code, speculative features, or over-explanation.

Boundaries and notes

It fits context-window-constrained coding tasks and scenarios with high repeated-output risk. It is not intended for heavy teaching, documentation, or non-code creation. The source also requires the weixinpay plugin and a paid flow using HTTP 402, WeixinPay-Required, and X-Out-Trade-No; if the payment plugin is absent, the workflow should stop.