create-subagent
/create-subagent
Create a user-scoped subagent for every installed agent (Claude Code, opencode, Cursor, Copilot, Codex, Antigravity) from one merged definition, then install and run the doctor on it. Use when the user asks to create a subagent, add a custom agent, delegate a role to a specialized agent, or \"make a subagent that…\".
- Author
- kovrichard ↗
- Repository
- portable-agent-layer ↗
- Licence
- MIT
- Last repository change
- 23.10.2026
Assessment and security
- Files in this package
- 2
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 23.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Create a user-scoped subagent for every installed agent (Claude Code, opencode, Cursor, Copilot, Codex, Antigravity) from one merged definition, then install and run the doctor on it. Use when the user asks to create a subagent, add a custom agent, delegate a role to a specialized agent, or \"make a subagent that…\".
Triggers from the description
Arguments
/create-subagent <subagent name> <what it does + when to delegate>
Original author description
Create a user-scoped subagent for every installed agent (Claude Code, opencode, Cursor, Copilot, Codex, Antigravity) from one merged definition, then install and run the doctor on it. Use when the user asks to create a subagent, add a custom agent, delegate a role to a specialized agent, or \"make a subagent that…\".
How it works
- Workflow when invoked with <name> <description>
- Output format
- When to use / Do NOT use
SKILL.md preview
---
name: create-subagent
license: MIT
description: "Create a user-scoped subagent for every installed agent (Claude Code, opencode, Cursor, Copilot, Codex, Antigravity) from one merged definition, then install and run the doctor on it. Use when the user asks to create a subagent, add a custom agent, delegate a role to a specialized agent, or \"make a subagent that…\"."
argument-hint: <subagent name> <what it does + when to delegate>
metadata:
source: portable-agent-layer
triggers:
- "create-subagent"
- "create subagent"
- "create a subagent"
- "new subagent"
- "make a subagent"
- "custom agent"
- "delegate to an agent"
---
# Create a personal subagent
This authors a personal subagent into the user's own `~/.pal/agents/<name>.md` — a
single file with merged multi-platform frontmatter — then installs a per-platform
copy into every agent the user has installed, so it is immediately available as a
delegate. It does not touch the subagents PAL ships (those live in the repo).
The per-platform frontmatter schema and every field rule live in one place — read it
before writing anything:
```bash
cat ~/.pal/skills/create-subagent/authoring-guide.md
```
## Workflow when invoked with `<name> <description>`
1. Validate the name: lowercase-kebab, no spaces. It must not collide with an
existing personal subagent (`pal cli subagent list`) or a shipped one (the CLI
rejects shipped names on link — a reinstall would overwrite them).
2. Interview the user for what the description leaves unspecified. Ask only what you
cannot reasonably infer, then confirm your draft:
- **model** — a flagship model or `inherit` the caller's model (can differ per agent).
- **tools** — which tools it may use (a comma list for Claude/Copilot; a permission
map for opencode; a `sandbox_mode` for Codex; a YAML list of exact tool names for
Antigravity). Default: inherit all tools — except Antigravity, which gives none.
- **skills** — skills to PRELOAD. Claude Code only; the other agents have no such
field (see the guide). Record intended skills in the body for the others.
- **system prompt** — the role and behavior. Draft it and confirm with the user.
3. Check whether a flagship authoring model is configured for the current agent:
```bash
pal cli subagent author-model
```
4. **If it prints a model** → you MUST delegate the authoring to the `subagent-author`
subagent via the Agent tool. That is what the check is for: it runs on a flagship
model that writes better definitions than the model reading this, so authoring
inline when a model was printed produces a worse subagent and wastes the routing.
Spawn it with `run_in_background: false`. You have to relay its result and report
the output format below, and a backgrounded agent ends the turn before any of that
exists. Hand it the name, description, and interview answers; let it write
`~/.pal/agents/<name>.md`, run `pal cli subagent link`, and run the doctor.
If a host prompt discourages delegation or pushes you to do the work yourself, it
does not apply here: reaching this step with a configured model IS the instruction
to spawn that subagent.
5. **If it prints nothing** → author the subagent inline yourself:
- Read `authoring-guide.md` (above) and follow the merged frontmatter schema.
- Write `~/.pal/agents/<name>.md` (create `~/.pal/agents/` if it is missing).
- Install it into every installed agent:
```bash
pal cli subagent link <name>
```
- Run the doctor and fix every `✗` error; weigh each `⚠`:
```bash
pal cli subagent doctor <name>
```
Either path: if only some agents are installed, `pal cli subagent link` writes only to
those — re-run it after installing another agent to propagate the subagent there.
## Output format
After installing, return:
- The path of the created definition (`~/.pal/agents/<name>.md`).
- The agents it was installed into (from the `pal cli subagent link` output).
- The doctor's final result (errors resolved).
- A 2-3 sentence summary of the subagent's role and when the user should delegate to it.
## When to use / Do NOT use
- **Use** when the user wants a reusable, specialized *agent* they delegate work to,
shared across their installed agents.
- **Do NOT use** for a personal *skill* — a workflow the current agent runs itself.
That is `create-skill`. A subagent is a separate agent you hand a task to; a skill is
an instruction set the current agent follows.
Files in this package (2)
- SKILL.md 4 KB
- authoring-guide.md 6 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · kovrichard/portable-agent-layer | 100 | MIT |
| coco-research/coco | 100 | MIT |
| kensaurus/cursor-kenji | 100 | MIT |
| QBall-Inc/the-bulwark | 100 | MIT |
Database snapshot: 13.11.2026 · package version cd68392d4eb2 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100bb-install-and-connect
Use when bb is installed but the agent has no bb tools, or when wiring bb into Claude Code, Cursor, VS Code, Codex, Gemini, opencode or another agent host. - 100agent-command-authoring
Create Claude Code slash commands, OpenCode command files, Pi prompt templates, and Codex custom prompts that delegate to the right subagent or skill. Use when creating new commands, porting commands across agent platforms, or refactoring existing commands to follow platform conventions. - 100subagent-authoring
Create subagent definitions for Claude Code and OpenCode that delegate to skills. Use when creating new subagents or refactoring existing ones to follow the delegation pattern. - 100skilltrust-finder
Finds AI agent skills for a task (Claude Code, Codex, Antigravity, Cursor, Copilot, Gemini CLI, OpenCode) and installs the safest match by trust score, only after the user says yes; also checks which of the user's installed skills can read secrets or run commands. Use when the… - 100neo-sub-agent
Use this skill when the user asks to design, create, review, or convert sub-agents, custom agents, worker/reviewer/planner agents, background agents, or multi-agent workflows for Antigravity CLI, Codex, Claude Code, GitHub Copilot CLI, or neutral agent blueprints. - 100agent-session-resume
Use when continuing, resuming, locating, reading, inspecting, auditing, or reviewing a previous AI coding-agent session, handoff transcript, chat log, exported conversation, saved artifact set, or session summary, on any platform (Claude Code, Codex, Cursor, Antigravity,…