Name: Prompt Subagent
Author: giladbarnea

name	prompt-subagent
description	Instructions for _how_ to prompt a subagent to perform any task. Load this skill before dispatching a subagent.

Orient the agent to the project: tell it to read key context files (README.md, AGENTS.md, ARCHITECTURE.md, etc.) and any relevant domain context directories—enough for it to understand the project’s purpose before starting the task. If the user referenced any context-gathering skills at the beginning of the main session, point the agent to them as well.
Be generous in giving the agent wider context—understanding why it's performing the task will boost its performance. Don't micromanage or over-instruct it. The agent already has a highly detailed system prompt. It is highly intelligent, just like you, and can navigate uncertainties well without being spoon-fed. Avoid prescribing instructions, giving "how-to" examples, providing examples as to what to think about, or dictating which files, symbols, or paths to look at; avoid any form of providing hints for possible answers for your own queries — this is circular and useless. Just declare what kind of understanding YOU are seeking for yourself. Instead of specifying which steps to take (dictating the "how" is bad), share only why it was dispatched and what you hope to achieve. This directly frees the agent to find the best way to reach your goal, unbiased and unconstrained by your own limited knowledge and assumptions. Research why Vercel claims their integrated version is beneficial (edge runtime, seamless DX, zero-config, billing, monitoring, tight coupling to vercel CLI / dashboard / functions). I want to know why Vercel claims their integrated version is beneficial.
Remind the agent that it can also spawn subagents to perform tasks, utilize skills, etc.
Subagents can take several minutes to run - use a 10-minute timeout.

name	prompt-subagent
description	Instructions for _how_ to prompt a subagent to perform any task. Load this skill before dispatching a subagent.

Orient the agent to the project: tell it to read key context files (README.md, AGENTS.md, ARCHITECTURE.md, etc.) and any relevant domain context directories—enough for it to understand the project’s purpose before starting the task. If the user referenced any context-gathering skills at the beginning of the main session, point the agent to them as well.
Be generous in giving the agent wider context—understanding why it's performing the task will boost its performance. Don't micromanage or over-instruct it. The agent already has a highly detailed system prompt. It is highly intelligent, just like you, and can navigate uncertainties well without being spoon-fed. Avoid prescribing instructions, giving "how-to" examples, providing examples as to what to think about, or dictating which files, symbols, or paths to look at; avoid any form of providing hints for possible answers for your own queries — this is circular and useless. Just declare what kind of understanding YOU are seeking for yourself. Instead of specifying which steps to take (dictating the "how" is bad), share only why it was dispatched and what you hope to achieve. This directly frees the agent to find the best way to reach your goal, unbiased and unconstrained by your own limited knowledge and assumptions. Research why Vercel claims their integrated version is beneficial (edge runtime, seamless DX, zero-config, billing, monitoring, tight coupling to vercel CLI / dashboard / functions). I want to know why Vercel claims their integrated version is beneficial.
Remind the agent that it can also spawn subagents to perform tasks, utilize skills, etc.
Subagents can take several minutes to run - use a 10-minute timeout.

prompt-subagent