code-work

Enforces scoped code reading and minimal editing conventions for code change tasks.

Updated Jul 11, 2026
One-click install
npx skills add https://github.com/danielsuguimoto/skills --skill code-work-danielsuguimoto
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: code-work
Source: https://github.com/danielsuguimoto/skills/tree/main/skills/code-work
Command: npx skills add https://github.com/danielsuguimoto/skills --skill code-work-danielsuguimoto

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? AI coding sessions often produce bloated diffs, unrequested refactors, speculative abstractions, and full-file rewrites that waste tokens and introduce risk. This Skill constrains code reading and editing to surgical, verifiable changes that match existing style. ## Core Features & Use Cases - Scoped Code Reading: Mandates symbol-level navigation and reference tracing before grep or full-file reads, with offset/limit reads for only the needed lines. - Surgical Editing Rules: Enforces minimal diffs, no code comments, no unrelated cleanup, and citation of existing code by path:lineRange instead of pasting bodies. - Goal-Driven Execution: Converts tasks into verifiable goals (e.g., write a failing test, then make it pass) with explicit step-by-step verification plans. - Use Case: When asked to fix a validation bug, the agent reads the relevant tests and symbol references first, edits only the broken function, removes only the imports its own change orphaned, and verifies via the test suite. ## Quick Start Apply the code-work conventions while fixing the validation bug in the user registration module.

Frequently Asked Questions about code-work

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make an AI coding agent produce minimal diffs?▼

Constrain the agent with surgical editing rules: touch only the code required for the change, match existing style, and never refactor adjacent code. Require it to state assumptions and define verifiable success criteria before editing.

How should an AI agent read large code files efficiently?▼

Use symbol-level code navigation and reference tracing first, then read with offset and limit for only the needed lines. Reserve full-file reads for files under roughly 150 lines or when the task genuinely requires the whole file.

When should grep be used instead of code navigation tools?▼

Use grep, read, or file search only for non-symbol lookups such as plain-text searches, known file paths, and glob patterns. Fall back to read when the navigation tool's LSP does not cover the language or the symbol is dynamic.

Should AI-generated code include comments?▼

Under these conventions, no code comments are written, overriding any provider default that encourages them. Existing code is referenced by path and line range instead of pasting bodies back into context.

What happens if the required code-navigation spec file is missing?▼

The agent falls back to native tools like grep and read, and explicitly notes the gap. It must never invent the contents of the missing specification file.