Agent-Skills.md

Agent Skills: brainstorming

Stress-test a plan or idea with grilling and domain modeling.

UncategorizedID: adibfirman/dotfiles/brainstorming

Repository

adibfirman/dotfiles
adibfirman
3

Install this agent skill to your local

pnpm dlx add-skill https://github.com/adibfirman/dotfiles/tree/HEAD/claude/.claude/skills/brainstorming

Skill Files

Browse the full folder contents for brainstorming.

Download Skill

Loading file tree…

claude/.claude/skills/brainstorming/SKILL.md

Skill Metadata

Name
brainstorming
Description
Stress-test a plan or idea with grilling and domain modeling.
  • Read grilling and domain-modeling. Follow both, except where this file says otherwise.
  • Never create or write any file, even when grilling or domain-modeling says to (for example GLOSSARY.md or an ADR). Show the text in chat first. Write the file only after the user says yes. This overrides both skills.
    • Always confirm to the user first.
  • Do not ask any question while a subagent is still running. This overrides the "Don't block on it" rule in grilling/SKILL.md.
    • Wait until every subagent finishes.
    • If a subagent runs for more than 5 minutes, tell the user which subagent is still running and what it searches for. Do not ask a question. Then keep waiting.
    • Keep every question, also the ones that an agent result answered. Mark an answered question with ✅, show the finding with a file:line link, and use the finding as the recommended answer. The user confirms or corrects it.
    • Ask the questions in one round, in the grilling format. Ask at most 5 open questions per round. ✅ questions do not count toward this limit. Ask first the open questions that unblock the most other questions. Keep the rest for the next round.
  • When a question is about a flow, a call path, or a dependency, add a diagram in the call-graph skill format. A diagram is often clearer than text alone.
  • Before each question round, check:
    • Am I anchoring on the first thing I found, or did I search broadly enough?
    • What did I expect to find but didn't?
    • What related code have I NOT read that could change my understanding? Hint: search with at least 3 different patterns, read callers AND callees, check tests for implicit contracts.
    • If a function comes from a dependency or a third-party package, check its docs with Context7 first. Use web docs only when Context7 has nothing.
  • At the end of the session, when the user confirms a shared understanding, post a short summary in chat: the decisions made, the open risks, and the next step. Include a recap of the final design in the call-graph skill format.
  • Put questions only at the end of the response, never in the middle of a topic or an explanation.
  • When a question has options, put each option on its own line, never in the same paragraph.