Skip to content

dimpletz/ai-engineer

v1.6.0MIT

AI Engineer is a collection of skills that help you create and optimize VS Code agents, skills, hooks, and plugins.

AI Engineer v1.6.0

A collection of skills for creating, optimizing, and auditing VS Code agents, skills, hooks, plugins, and custom AI instruction files.

Prerequisites

Installation

Install via the VS Code Chat Plugin Marketplace using the dimpletz/prompts-collection marketplace source and enable the ai-engineer plugin.

Usage

Most capabilities are provided as skills — invoke them by describing your intent in Copilot Chat. Copilot will automatically select the appropriate skill when the request matches. This plugin also includes lifecycle hooks that inject context guidance at session and subagent start.

SkillInvoke when…
Agent MakerYou want to create a new .agent.md file for a custom VS Code agent, either at workspace level (.github/agents/) or inside a specific plugin directory.
Agent OptimizerYou want to improve, refactor, or decompose an existing .agent.md into an orchestrator + subagent architecture.
Skill MakerYou want to design a new SKILL.md file for an AI skill, either at workspace level (.agents/skills/) or inside a specific plugin directory.
Custom Instruction MakerYou want to create an AGENTS.md, copilot-instructions.md, or CLAUDE.md instruction file that follows the Purpose → Tree → Rules structure.
Hook MakerYou want to create a lifecycle hook (hooks.json + scripts) that runs automatically at session start, subagent start, or after tool use.
Plugin MakerYou want to scaffold a new plugin with a plugin.json manifest, README, and the correct directory structure. Always places plugin.json at the plugin root.
Marketplace MakerYou want to register or synchronize the project's plugins in the VS Code Chat Plugin Marketplace (marketplace.json + README.md).
Instruction AuditorYou want a report-only review of any Markdown file that instructs a model (AGENTS.md, copilot-instructions.md, .agent.md, SKILL.md, etc.) for contradictions, repetition, ambiguity, safety issues, stale references, and missing coverage.

Components

graph TD
    A[ai-engineer plugin]
    A --> B[Agent Maker\nskills/agent-maker/SKILL.md]
    A --> C[Agent Optimizer\nskills/agent-optimizer/SKILL.md]
    A --> D[Skill Maker\nskills/skill-maker/SKILL.md]
    A --> E[Custom Instruction Maker\nskills/custom-instruction-maker/SKILL.md]
    A --> F[Hook Maker\nskills/hook-maker/SKILL.md]
    A --> G[Plugin Maker\nskills/plugin-maker/SKILL.md]
    A --> H[Marketplace Maker\nskills/marketplace-maker/SKILL.md]
    A --> J[Instruction Auditor\nskills/instruction-auditor/SKILL.md]
    A --> I[Temporary Script Guidance Hook\ncom.github.copilot/hooks/hooks.json]

Agent Maker

Creates well-structured, production-ready VS Code agent files (.agent.md) with required frontmatter fields and mandatory body sections. Saves to plugins/<plugin-name>/com.github.copilot/agents/ when a plugin name is provided, or to .github/agents/ at workspace level. Enforces consistent structure while allowing domain-specific flexibility.

Agent Optimizer

Analyzes and improves existing VS Code agent files. Covers quality analysis, guardrail strengthening, workflow optimization, and decomposing monolithic agents into orchestrator + subagent architectures. Preserves the original file's scope — saves optimized files in the same directory as the source (plugin or workspace).

Skill Maker

A meta-skill for designing new SKILL.md files. Clarifies purpose, inputs, priorities, and workflow so that downstream assistants behave predictably and are easy to maintain. Saves to plugins/<plugin-name>/skills/<skill-name>/SKILL.md when a plugin name is provided, or to .agents/skills/<skill-name>/SKILL.md at workspace level.

Custom Instruction Maker

Creates structured AI instruction files (AGENTS.md, copilot-instructions.md, CLAUDE.md) following the Purpose → Tree → Rules structure with a built-in self-improving note-taking engine.

Hook Maker

Creates VS Code Copilot agent hook configurations (hooks.json and companion shell/PowerShell scripts). Covers all lifecycle events (SessionStart, SubagentStart, PostToolUse), both context-injection and exit-code-enforcement patterns, and cross-platform script generation.

Plugin Maker

Scaffolds a complete VS Code Copilot agent plugin: the plugin.json manifest (always at the plugin root), directory structure, and README.md. Also registers the plugin in marketplace.json.

Marketplace Maker

Registers, synchronizes, and updates the project's plugins in the VS Code Chat Plugin Marketplace. Keeps .github/plugin/marketplace.json and the global README.md (Plugins table, Agents catalog, Skills catalog, Hooks table) in sync with the actual plugin contents.

Instruction Auditor

Audits any Markdown file that instructs an AI model and produces an evidence-backed report: contradictions (including cross-file and frontmatter-versus-body), redundancy, ambiguity, safety and prompt-injection risks, stale paths and commands, length and structure, file-type fitness, and missing coverage. Every finding cites a quote and line number and carries a severity. Report-only — it never edits the audited file. The full report is always written to a Markdown file (default reports/<file-name>-audit-<timestamp>.md at the workspace root) and only the file path is returned in chat.

Hooks

Runs on SessionStart and SubagentStart to inject guidance to use direct shell/PowerShell commands for large, simple file handling (for example, searching text), and use temporary shell/PowerShell scripts for large, complex file handling, then remove temporary scripts after use.

Author

Dimpletz