new-agent
Use when the user wants to create a specialized agent for Antigravity, Codex, ChatGPT, Zed, OpenCode, Qoder, Orca, Cursor, GitHub Copilot / VS Code, Windsurf…
Skill by Starraider
Chalk Labs
Need help deploying new-agent?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/starraider-new-agent
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Nov 1, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/new-agent
What's inside
- Skill name
new-agent- Description
- Use when the user wants to create a specialized agent for Antigravity, Codex, ChatGPT, Zed, OpenCode, Qoder, Orca, Cursor, GitHub Copilot / VS Code, Windsurf, Kiro, or Google Agents CLI. Require the target product and the agent's specific capabilities, then create the smallest supported agent definition and give product-appropriate verification steps. Do not treat skills, rules, or generic prompts as agents unless the selected product documents that model.
- License
- CC-BY-4.0
- Instructions
- 83 lines in SKILL.md · view on GitHub
Before installing
- Instructions read files in your home folder (~/.ssh, ~/.aws…)
Files (14)
- README.md
- SKILL.md
- evals.json · evals/
- target-formats.md · references/
- antigravity-agent.md · templates/
- codex-agent.toml · templates/
- copilot-agent.md · templates/
- cursor-mode.mdc · templates/
- google-agents-cli-spec.md · templates/
- kiro-agent.json · templates/
- opencode-agent.md · templates/
- qoder-agent.md · templates/
- windsurf-rules.md · templates/
- zed-agent-settings.jsonc · templates/
SKILL.md (first lines)
# New Agent ## Outcome Create a clearly scoped, least-privilege agent for exactly one requested target: Antigravity, Codex, ChatGPT, Zed, OpenCode, Qoder, Orca, Cursor, GitHub Copilot / VS Code, Windsurf, Kiro, or Google Agents CLI. Produce the native configuration or a ready-to-paste builder specification, plus discovery and smoke-test instructions. Do not create, publish, or enable a remote ChatGPT agent or a Google Agents CLI cloud deployment without the user's explicit approval. ## Workflow 1. Establish the agent contract before writing a definition. Require both of the following. If either is missing, ask one concise clarification question rather than choosing on the user's behalf: - **Target IDE/product:** Antigravity, Codex, ChatGPT, Zed, OpenCode, Qoder, Orca, Cursor, GitHub Copilot / VS Code, Windsurf, Kiro, or Google Agents CLI. If the user names a product with multiple agent surfaces, identify the requested one. - **Specific capabilities:** the agent's single responsibility, expected inputs and deliverable, tools or integrations it needs, and whether it may edit files, run commands, or make external writes. Also resolve a name, project versus personal scope, model preference if material, and whether the agent is a main agent, profile, subagent, or external command. Explain any unsupported capability or missing account permission before generating files. Completion: the purpose, target surface, scope, and least-privilege tool set are unambiguous. 2. Select the target's documented agent representation. Read [target formats and research](references/target-formats.md). Use only the branch for the selected product: - **Antigravity:** create a Markdown custom agent at `.agents/agents/<name>/agent.md` for a workspace or `~/.gemini/config/agents/<name>/agent.md` for a user. Use the Antigravity template. - **Codex:** when the supported local runtime exposes custom agent definitions, create `.codex/agents/<name>.toml` (or `~/.codex/agents/<name>.toml`) from the Codex template. Otherwise, do not falsely substitute a Skill: offer a ChatGPT Workspace Agent only if its workspace plugin is available, or explain that the current Codex surface cannot load the requested local agent. - **ChatGPT / ChatGPT Desktop App:** distinguish a Custom GPT from a Workspace Agent. Explain that both are authored through OpenAI's web builder / Agent Studio interfaces (not local desktop configuration files) and can be accessed in the ChatGPT Desktop App once created. Generate a complete builder brief—name, description, instructions, starters, knowledge, tools, least-privilege action constraints, and test prompts—then ask for confirmation before creating or publishing it in the browser or Workspace Agents app. - **Zed:** distinguish a native Zed Agent profile (tools/model for the Zed Agent) from an ACP external agent process. Generate the applicable `settings.json` fragment; use the Zed profile template or an `agent_servers` command entry. Do not imply an ACP entry creates a new LLM agent implementation. - **OpenCode:** create `.opencode/agents/<name>.md` for project scope or `~/.config/opencode/agents/<name>.md` for user scope. Use the OpenCode template and explicit permissions. - **Qoder:** create `.qoder/agents/<name>.md` for project scope or `~/.qoder/agents/<name>.md` for user scope. Use the Qoder template; include only tools, skills, and MCP servers the user requires. - **Orca:** create an Orca custom CLI-agent launch specification: name, verified binary or command, arguments, and optional safe startup hook. Orca registers a CLI agent in its Settings UI; it does not define the agent's internal behavior, so create the underlying CLI agent configuration first if one is needed. - **Cursor:** distinguish a Custom Mode (persona with tool permissions, UI-configured) from a project rule (`.cursor/rules/<name>.mdc`, version-controlled). Generate a Custom Mode specification for a persona or role request; generate a `.mdc` file for a project-knowledge request. Explain that Custom Modes are not version-controlled by default. Use the Cursor mode template. - **GitHub Copilot / VS Code:** create `.github/agents/<name>.agent.md` at repository scope, or the equivalent user-profile path for cross-workspace use. Distinguish the `.agent.md` agent from the always-on `copilot-instructions.md` and from path-scoped `.instructions.md` files; do not substitute them. Offer a Copilot Extension brief only when the user explicitly requires custom runtime behavior beyond what `.agent.md` supports. Use the Copilot agent template. - **Windsurf:** explain that Windsurf does not have a discrete named-agent file format. The closest supported equivalent is a named rule file inside `.windsurf/rules/<name>.md` with Always On activation, using XML-style tags to group role, operating rules, and boundaries. Distinguish project-level (`.windsurfrules` or `.windsurf/rules/`), global (`global_rules.md`), and Cascade Memories. Alert the user to review Memories when updating rules. Use the Windsurf rules template. - **Kiro:** create `.kiro/agents/<name>.json` (or `.md`) for project scope or `~/.kiro/agents/<name>.json` for user scope. Distinguish custom agents from steering files (`.kiro/steering/`) and hooks (`.kiro/hooks/`). Start with `tools: ["read"]` only; add write or shell only on request. Include steering files via `resources` rather than embedding their content. Use the Kiro agent template. - **Google Agents CLI:** distinguish an ADK agent project (Python code in `agent.py` scaffolded by `agents-cli create`) from the agents-cli skills injected into a coding assistant. Produce a design specification using the Google Agents CLI spec template, then scaffold the project. Do not claim a Markdown spec alone is the agent. Ask before triggering `agents-cli deploy` or publishing to the Gemini Enterprise Agent Registry. Completion: exactly one native representation is selected, and the output does not claim portability across IDEs. 3. Write a focused system instruction and capabilities boundary. Write the body prompt for the selected agent, not a generic persona. State its responsibility, inputs to inspect, ordered operating rules, expected deliverable, verification required, and stop/return conditions. Keep selection metadata (name and description) separate from operating instructions. Start read-only. Add file writes, shell access, web access, MCP servers, apps, actions, or other external-write capability only when the user explicitly requires it. For any external write, require confirmation immediately before the action unless the target has a stricter documented approval control. Completion: every permission maps to a stated capability; no enabled tool is speculative. 4. Create the artifact and keep side effects within scope. For local IDE definitions, create only the chosen file and parent directory. Preserve an existing same-named definition unless the user explicitly asked to update it. For ChatGPT or Orca UI definitions, present the configuration and the exact creation steps; use an available native connector or UI only after the user has authorized the remote/UI change. Completion: the artifact is at the requested scope, contains no secrets, and has no unrelated configuration changes. 5. Verify discovery and behavior. Follow the selected target's validation steps in [target formats and research](references/target-formats.md). Verify discovery first, then invoke the agent explicitly with one representative request. Confirm that a read-only agent does not gain edit or command access and that any expected integration is both configured and authorized. Completion: the agent is visible in the target UI/CLI, handles its representative task, and respects the declared tool boundary; otherwise report the exact discovery or permission blocker. ## Safety - Never invent a client-specific agent schema or claim that an `AGENTS.md`, a Rule, a Skill, or an MCP server is itself a custom agent. - Do not store API keys, OAuth tokens, or passwords in agent files, startup hooks, prompts, templates, or generated commands. - Treat apps, actions, MCP servers, shell access, and write permissions as capability grants, not as instructions. Keep them least-privilege and subject to the target's approval system. - ChatGPT creation, sharing, publishing, schedules, API triggers, and connector changes are external state changes. Obtain explicit user authority for each requested action. - If the request has no specific capability or no selected target product, stop and ask for it. Do not create a generic cross-IDE agent. ## Resources - If it is unclear whether the capability needs a specialist agent, use [`improve-my-ai-harness`](../improve-my-ai-harness/README.md) before creating a definition. - For reusable instructions without a dedicated tool boundary, use [`new-skill`](../new-skill/README.md); for repository-wide guidance, use [`new-agents-md`](../new-agents-md/README.md). - [Target formats, validation, and source research](references/target-formats.md) - [Antigravity Markdown agent template](templates/antigravity-agent.md) - [Codex TOML agent template](templates/codex-agent.toml) - [OpenCode Markdown agent template](templates/opencode-agent.md) - [Qoder Markdown agent template](templates/qoder-agent.md) - [Zed settings templates](templates/zed-agent-settings.jsonc) - [Cursor Custom Mode / rule template](templates/cursor-mode.mdc) - [GitHub Copilot agent template](templates/copilot-agent.md) - [Windsurf rules template](templates/windsurf-rules.md) - [Kiro agent JSON template](templates/kiro-agent.json) - [Google Agents CLI spec template](templates/google-agents-cli-spec.md)
From Starraider/ai-skills-plugin, CC-BY-4.0.
More from Starraider/ai-skills-plugin
- 0.603score
improve-my-ai-harness
Starraider
Diagnose how to improve an AI coding IDE or agent harness by interviewing the user, identifying required capabilities, selecting a prompt, AGENTS.md, Agent…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-slash-command
Starraider
Use when deciding whether a reusable task should be a slash command and when creating or revising a target-native command or the documented alternative for…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-workflow
Starraider
Use when deciding whether a repeatable workflow is appropriate and when creating or revising a target-native workflow for Antigravity, Codex, ChatGPT, Zed…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-hook
Starraider
Use when deciding whether a lifecycle or event hook is the right solution, or when creating or revising one for Antigravity, Codex, ChatGPT, Zed, OpenCode…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-plugin
Starraider
Create, package, migrate, release, or validate portable Agent Plugins 1.0.0 containing Agent Skills and optional MCP servers.
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-prompt
Starraider
Create or improve a ready-to-use LLM prompt from a user's goal.
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-skill
Starraider
Use when creating, improving, validating, or packaging an Agent Skill.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
improve-skill
Starraider
Review and propose general-purpose updates to skills used in the current conversation.
SkillAFreshness: A★ 2Claude Code +9 - 0.578score
new-agents-md
Starraider
Use when the user wants to create, replace, or improve an AGENTS.md instruction file for Antigravity, Codex, ChatGPT, Zed, OpenCode, Qoder, or Orca.
SkillAFreshness: A★ 2Claude Code +9
Signals
- Stars
- 2
- Forks
- 1
- Open issues
- 29
- Last commit (this folder)
- Sep 24, 2026 (1 month ago)
- Last release
- Sep 26, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- CC-BY-4.0
- Repository
- github.com/Starraider/ai-skills-plugin
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from Starraider
- 0.603score
improve-my-ai-harness
Starraider
Diagnose how to improve an AI coding IDE or agent harness by interviewing the user, identifying required capabilities, selecting a prompt, AGENTS.md, Agent…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-slash-command
Starraider
Use when deciding whether a reusable task should be a slash command and when creating or revising a target-native command or the documented alternative for…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-workflow
Starraider
Use when deciding whether a repeatable workflow is appropriate and when creating or revising a target-native workflow for Antigravity, Codex, ChatGPT, Zed…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-hook
Starraider
Use when deciding whether a lifecycle or event hook is the right solution, or when creating or revising one for Antigravity, Codex, ChatGPT, Zed, OpenCode…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-plugin
Starraider
Create, package, migrate, release, or validate portable Agent Plugins 1.0.0 containing Agent Skills and optional MCP servers.
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
new-prompt
Starraider
Create or improve a ready-to-use LLM prompt from a user's goal.
SkillAFreshness: A★ 2Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get new-agent running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
new-agent skill. Agent Mart. https://agentmart.dev/skills/starraider-new-agent (last updated 2026-10-24)