skills.rest
liuzhengdongfortest
liuzhengdongfortestCOMMUNITY
Explore all skills by this author →
GitHub

build-cs-skill

Creates and evolves thin-harness CodeStable skills with explicit rule placement and evidence gates.

1.1k|83|Updated Apr 12, 2026
Search Tags:#agent workflow#Prompt Engineering#codestable#Quality gates#SKILL authoring#rule placement#haskell contracts
One-click install
npx skills add https://github.com/liuzhengdongfortest/easysdd --skill build-cs-skill
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: build-cs-skill
Source: https://github.com/liuzhengdongfortest/easysdd/tree/main/.claude/skills/build-cs-skill
Command: npx skills add https://github.com/liuzhengdongfortest/easysdd --skill build-cs-skill

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Authoring AI agent skills often produces bloated prompt files that mix responsibilities, duplicate rules, and lack verifiable completion criteria. This Skill provides a structured protocol for creating, refactoring, simplifying, and reviewing cs-* skills so each one stays a minimal control surface backed by on-demand context and testable evidence gates.

Core Features & Use Cases

  • Shape Selection: Classifies each skill into a typed shape (ThinOperator, ContextualWorkflow, ToolBackedWorkflow, ShimSkill, ReferenceSkill, or NoActiveSkill) using Haskell-style decision contracts before any writing begins.
  • Rule Placement: Routes every rule to exactly one canonical home (harness, stage context, project context, deterministic gate, or removal) to eliminate duplication and additive-only bloat.
  • Quality Gates & Fixtures: Ships reference standards for review algorithms, regression ladders, runtime conformance testing, and risky routing/checkpoint fixture patterns.
  • Use Case: When a new cs-* skill is needed or an existing one has grown unwieldy, invoke this Skill to reclassify its shape, relocate misplaced rules, build a context plan, and produce validation evidence.

Quick Start

Use the build-cs-skill to refactor the cs-feat skill under plugins/codestable/skills into a thin harness with an explicit context plan and evidence gates.

Frequently Asked Questions about build-cs-skill

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

FAQPage Schema
How do I create a new CodeStable skill with a thin harness?▼

Define the skill's single responsibility, classify its shape with selectSkillShape, place every rule via placeRule, build a context plan, then write the harness in the prescribed order: trigger, responsibility, context contract, gates, and done/failure conditions.

How do I refactor an existing cs-* skill that has grown too large?▼

Inventory all current rules, re-run each through placeRule, move stage methods to references and project facts to .codestable artifacts, mechanize deterministic checks, and delete prose that no longer changes decisions, gates, or recovery.

When should a skill use a Haskell contract versus plain prose?▼

Use Haskell-style contracts only for closed decisions, transitions, lifecycles, or invariants with typed boundaries and guard-ordered equations. Simple operators with a few guards should stay as plain prose without a state machine.

What is the difference between ThinOperator and ContextualWorkflow shapes?▼

ThinOperator fits a single action or artifact with minimal guards and no recoverable state. ContextualWorkflow applies when the skill has real stages, recovery, iteration, or owner checkpoints that require loading context per stage.

When should I not use build-cs-skill?▼

Do not use it for product feature implementation, which belongs to cs-feat or cs-issue. Also avoid it for measured multi-model experiment loops, which are owned by eval-cs-skill, and never recreate retired v1 skill names as shims.

Related Skills

Explore More Software Engineering Skills→
TakfesTakfes

skill-transfer

Compare skill directories and report duplicates, conflicts, and transferable updates.

Community
Intermediate
CDO-07-NewCDO-07-New

eval-harness

Define repeatable evaluation harnesses for Claude Code task outcomes.

Official
Intermediate
nsalvacaonsalvacao

memory-management

Decode shorthand and internal terms using CLAUDE.md and memory/ files.

Community
Advanced
stlwolfstlwolf
2

implementer-contract

Define a structured return contract for subagent task delegation.

Community
Intermediate