AGENT SKILL

subagent-grilling

Grill a strong-model subagent down a plan's whole design tree in the user's place, escalating only what needs the user. Use when the user wants a stronger model to answer a plan's design questions.

License 许可
MIT
最近更新
2026-11-13
Stars
★ 2
Forks
0

01 前言 Front Matter

name
subagent-grilling
description
Grill a strong-model subagent down a plan's whole design tree in the user's place, escalating only what needs the user. Use when the user wants a stronger model to answer a plan's design questions.

03 SKILL.md 正文

Run the grilling skill with a subagent, the interviewee, in the user's seat: you ask, it answers. A strong model makes the calls, you keep the interview relentless, and the user hears only what the escalating-to-user skill escalates. Both skills hold except where this skill overrides them.

Confirm the interviewee's model

Before spawning, confirm the interviewee's model with the user, unless they already named one or the harness offers no choice. List the available models or configured roles, recommend the strongest at design reasoning with high effort, and word the question so "yes" accepts. Spawn nothing until they answer.

Brief the interviewee

Spawn a persistent subagent you can send round after round. It shares none of your context, so the brief carries:

  • The goal and the plan as it stands, with the decisions and constraints the user has already settled.
  • Paths to the files, docs, and code it bears on.
  • Its role: answering a design interview on the user's behalf under the escalating-to-user skill. Answer every question in a round, numbered to match, with a decision, a one-line why, and the strongest alternative rejected. Flag an answer that passes the escalation tests as ESCALATE, keeping its recommendation. Flag an answer whose load-bearing decision rests on a public fact as RESEARCH, naming the fact it assumed and leaving the lookup to the asker.

Run the rounds

Send each round to the interviewee in place of the user. Challenge an answer that is thin or clashes with what the user has said; the challenge joins the next round.

Settle each RESEARCH flag, and any public fact you spot a load-bearing decision resting on, with the research skill, and put its findings to the interviewee in the next round; the findings file is the decision's artifact.

The escalation call is yours, since you hold the user's context: run each answer through escalating-to-user's tests, whether flagged or not. An escalated answer becomes your provisional recommendation; tell the interviewee it is provisional.

Present the ledger

When the frontier is empty, every node is decided or escalated. Present the ledger with the design tree as its decisions. Send the user's answers and overrides to the interviewee as settled, and reopen the subtrees they change in new rounds.

04 同仓库更多 Skills

ian-pascoe
ian-pascoe/dotfiles

agent-browser

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.

测试与质量Skill与智能体开发
★ 2
ian-pascoe
ian-pascoe/dotfiles

writing-for-agents

Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.

内容创作者写作与文案
★ 2
ian-pascoe
ian-pascoe/dotfiles

wizard

Generate an interactive bash wizard that walks a human through steps only they can perform. Use when provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover. Don't invoke this for steps the agent can perform itself.

★ 2
ian-pascoe
ian-pascoe/dotfiles

wayfinder

Plan a huge chunk of work (more than one agent session can hold) as a shared map of decision tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.

★ 2
ian-pascoe
ian-pascoe/dotfiles

wait-what

Stop. That last message did not land: re-pitch it.

★ 2
ian-pascoe
ian-pascoe/dotfiles

unslop

Cut AI tells from any writing.

★ 2