agent-development
Use this skill when the user asks to create an agent, add a subagent, write a system prompt, define triggering conditions, choose tools or model settings, or adapt an agent format for a different client.
Best use case
agent-development is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Use this skill when the user asks to create an agent, add a subagent, write a system prompt, define triggering conditions, choose tools or model settings, or adapt an agent format for a different client.
Teams using agent-development should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/skill/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How agent-development Compares
| Feature / Agent | agent-development | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Use this skill when the user asks to create an agent, add a subagent, write a system prompt, define triggering conditions, choose tools or model settings, or adapt an agent format for a different client.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Agent Development Use this skill to design agents that handle complex, multi-step work with a clear role, prompt, and trigger logic. ## Overview Agents are autonomous subprocesses for structured work. Commands are for user-initiated actions. A good agent has three parts: 1. A narrow role 2. Clear trigger conditions 3. A system prompt that defines behavior and output ## This repo's format In this Nix setup, agent definitions are usually split across: 1. `modules/ai/clients/agents/prompts/<agent>.md` for the system prompt 2. `modules/ai/clients/agents/default.nix` for metadata like model, color, temperature, tools, and prompt path If you are targeting another client, map the same concepts to that client's agent file format. ## Agent structure ### Portable metadata When a client supports agent frontmatter, the equivalent shape is: ```markdown --- name: agent-name description: Use this agent when ... Examples: color: blue tools: ["Read", "Write", "Grep"] --- ``` ### System prompt Write the prompt in second person and keep it specific: ```markdown You are a [role] specializing in [domain]. **Your Core Responsibilities:** 1. ... 2. ... **Analysis Process:** 1. ... 2. ... **Output Format:** ... ``` ## Frontmatter guidance ### Name Use a lowercase, hyphenated identifier. ### Description Use the description to define when the agent should trigger. Include 2 to 4 concrete examples and cover alternate phrasings of the same intent. ### Model Prefer the smallest capable model for the job. Use the default model when the client supports inheritance. ### Color Use color as a quick visual hint for the agent's role. ### Tools Limit tools to the minimum required. Read-only agents should not get write access. ## Prompt design Include these parts when useful: 1. Responsibilities 2. Analysis process 3. Quality standards 4. Output format 5. Edge cases Keep the prompt direct, testable, and specific to the agent's job. ## Creating agents ### AI-assisted generation Use a prompt like this: ```text Create an agent configuration for: "[request]" Return: - identifier - whenToUse - systemPrompt - model - color - tools ``` Then adapt the result to the target client's format. ### Manual creation 1. Choose the agent's narrow job 2. Write trigger examples 3. Draft the system prompt 4. Pick model, color, and tools 5. Wire it into the client's agent configuration ## Validation rules 1. The agent name should be unambiguous 2. The description should make triggering obvious 3. The prompt should define output format 4. The tool set should be minimal 5. The agent should fail closed on unclear requests ## Implementation workflow in this repo 1. Add the prompt file under `modules/ai/clients/agents/prompts/` 2. Add or update the agent metadata in `modules/ai/clients/agents/default.nix` 3. Keep the prompt portable so it can be reused by multiple clients 4. Verify the agent trigger wording matches the intended use Focus on clarity over cleverness. A good agent is specific enough to be useful and narrow enough to be reliable. ## References - `references/system-prompt-design.md` - `references/triggering-examples.md` - `references/agent-creation-system-prompt.md`
Related Skills
repo-practicing-trunk-based-development
Trunk Based Development workflow - all development on main branch with small frequent commits, minimal branching, and continuous integration. Covers when branches are justified (exceptional cases only), commit patterns, feature flag usage for incomplete work, environment branch rules (deployment only), and AI agent default behavior (assume main). Essential for understanding repository git workflow and preventing unnecessary branch proliferation
performing-yara-rule-development-for-detection
通过识别可执行文件中的唯一字节模式、字符串和行为指标，开发精准的 YARA 恶意软件检测规则，同时将误报率降至最低。
wp-plugin-development
Use when developing WordPress plugins: architecture and hooks, activation/deactivation/uninstall, admin UI and Settings API, data storage, cron/tasks, security (nonces/capabilities/sanitization/escaping), and release packaging.
wp-block-development
Use when developing WordPress (Gutenberg) blocks: block.json metadata, register_block_type(_from_metadata), attributes/serialization, supports, dynamic rendering (render.php/render_callback), deprecations/migrations, viewScript vs viewScriptModule, and @wordpress/scripts/@wordpress/create-block build and test workflows.
iterative-development
Implement sub-tasks iteratively with user approval between each one. Use when working through a task list with user collaboration.
doubt-driven-development
Use when a non-trivial decision should not stand without a fresh-context challenge - isolate the artifact and contract, run an adversarial review before committing, and reconcile the findings while change is still cheap.
spec-driven-development
Use before starting any non-trivial feature to write a technical spec first — prevents misaligned implementation and scope creep
source-driven-development
Use when implementing framework-specific or library-specific code — verify non-trivial APIs against official documentation before writing them, and record the source in task notes, docs, or the PR.
using-development-skills
Use when starting any conversation - establishes how the development-skills plugin works and how to invoke its components on each platform (Claude Code, Codex). Read first.
Finishing a Development Branch
Use this when you have completed some feature implementation and have written passing tests, and you are ready to create a PR.
lwr-site-development
Use this skill when building or customizing sites on the Lightning Web Runtime (LWR) in Experience Cloud — including component authoring, custom theming with --dxp hooks, layout components, and publish lifecycle management. Trigger keywords: build LWR site Experience Cloud, Lightning Web Runtime custom theme, LWR component development community, Build Your Own LWR template, Microsite LWR, lightningCommunity__Theme_Layout, --dxp styling hooks. NOT for Aura-based communities (Build Your Own Aura template). NOT for standard Experience Builder drag-and-drop configuration without code.
data-cloud-activation-development
Use this skill when building developer-driven Data Cloud activation surfaces: webhook Data Action Targets with HMAC-SHA256 signing, Salesforce Platform Event data actions, Data Cloud-Triggered Flows on DMO insert, or Marketing Cloud journey triggers. Triggers on: webhook data action target, Data Cloud triggered Flow not firing, HMAC secret key for data action, platform event from Data Cloud, DMO insert trigger. NOT for configuring standard admin-level Activation Targets (SFTP, ad platform segment publishing, CRM segment activation) — those require admin configuration skills, not this developer extensibility skill.