Help us improve
Share bugs, ideas, or general feedback.
From electrified-cortex-skills
How to dispatch a sub-agent. Triggers - run in background, spawn subagent, background task, isolated agent execution, dispatch subagent, background agent.
How this skill is triggered — by the user, by Claude, or both
Slash command
/electrified-cortex-skills:dispatchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Input:
Share bugs, ideas, or general feedback.
Input:
<prompt> — verbatim prompt sent to sub-agent
<description> — short run label shown by host
<tier> — fast-cheap | standard (default) | deep
<model-override> (optional) — concrete model or alias (e.g. Claude Sonnet 4.6, GPT-5.4, gpt-5-codex); bypasses tier lookup when set
Derived:
<concrete-model> = <model-override> if set, else derived from <tier> via table.
Process:
If <prompt> instructs sub-agent to read a file, don't read it yourself — sub-agent does. Spawn zero-context Dispatch sub-agent.
DO NOT DISPATCH SKILLS — read them. The skill itself tells you when (and how) to dispatch.
Claude Code:
Claude Model Aliases:
| Tier | Class | model value |
|---|---|---|
fast-cheap | haiku-class | haiku |
standard | sonnet-class | sonnet |
deep | opus-class | opus |
Agent({
subagent_type: "Dispatch",
prompt: "<prompt>",
model: "<concrete-model>",
run_in_background: true,
description: "<description>"
})
VS Code / Copilot:
Copilot Model Aliases:
| Tier | Class | model value |
|---|---|---|
fast-cheap | haiku-class | Claude Haiku 4.5 |
standard | sonnet-class | Claude Sonnet 4.6 |
deep | opus-class | Claude Opus 4.6 |
GPT alts (gpt-class): GPT-5.3-Codex (code), GPT-5.4 (prose), GPT-5.4 mini (fast-cheap prose).
Update minimum models as needed.
runSubagent({
agentName: "Dispatch",
prompt: "<prompt>",
model: "<concrete-model>",
description: "<description>"
})
Fallback:
If the "Dispatch" agent is not installed: omit subagent_type / agentName and continue — behavior is identical. The agent adds context isolation and consistent performance. Notify the host after completion.
If the requested model is not available: stop and inform the caller; suggest an alternative.
Return: Return (passthrough) sub-agent output to caller.
Concurrency: When dispatching multiple instructions simultaneously, default max concurrent = 3. Use a rolling window: start up to 3, then as each returns dispatch the next pending instruction. Only exceed 3 if instructed by the caller.
See also:
supplemental.md — context inheritance, CLI dispatch, hash-record. dispatch-pattern.md — design rationale. installation.md — agent install. setup/SKILL.md — VS Code/Cursor setup.
npx claudepluginhub electrified-cortex/skills-plugin --plugin electrified-cortex-skillsCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.