Agent-Skills.md

Agent Skills: brainstorming

Proactively use when brainstorming potential solutions to difficult problems.

UncategorizedID: hdemers/dotfiles/brainstorming

Repository

hdemers/dotfiles
hdemers
31

Install this agent skill to your local

pnpm dlx add-skill https://github.com/hdemers/dotfiles/tree/HEAD/.agents/skills.symlink/brainstorming

Skill Files

Browse the full folder contents for brainstorming.

Download Skill

Loading file tree…

.agents/skills.symlink/brainstorming/SKILL.md

Skill Metadata

Name
brainstorm
Description
Proactively use when brainstorming potential solutions to difficult problems.

Standard Operating Procedure

MANDATORY: Use the report-writing skill.

Phase 1: Low-Hanging Fruits

  1. Generate 4-5 solutions
  2. Do not assume backward compatibility, we want to explore the full space of solutions.
  3. Follow all best practices below
  4. STOP and ASK: validate solutions with user. DO NOT move to phase 2.
  5. If user selects a solution, do not start implementing. Always ask the user: should we make a plan first. Then, make a plan.

Phase 2: Advanced Solutions

  1. Only engage in phase 2 if user wants more advances solutions.
  2. Do not assume backward compatibility, we want to explore the full space of solutions.
  3. Launch 4-5 subagents each tasked with a slightly different possible solution.
  4. Subagent tasks:
    1. Each subagents shall test their proposed solution with a throw-away script demonstrating the validity of their approach.
    2. Subagents shall not stop until they have a valid solution.
    3. Subagents shall follow the best practices found below.
  5. Summarize and rank each solutions
  6. STOP and ASK: validate solutions with user. DO NOT move to phase 3.
  7. If user selects a solution, do not start implementing. Always ask the user: should we make a plan first. Then, make a plan.

Phase 3: Randomized Search Over Solution Space

  1. Only engage in phase 3 if user wants more solutions.
  2. Do not assume backward compatibility, we want to explore the full space of solutions.
  3. Launch 4-5 subagents each tasked with the exact same ask/prompt.
  4. Subagent tasks:
    1. Each subagents shall test their proposed solution with a throw-away script demonstrating the validity of their approach.
    2. Subagents shall not stop until they have a valid solution.
    3. Subagents shall follow the best practices found below.
  5. Summarize and rank each solutions
  6. STOP and ASK: validate solutions with user. DO NOT move to phase 3.
  7. If user selects a solution, do not start implementing. Always ask the user: should we make a plan first. Then, make a plan.

Best Practices

  1. Implementing a solution is not the goal, rather writing a report is.

  2. When evolving a solution, amend the existing report instead of writing a new one.

  3. Complex solutions should be validated with throw-away scripts.

  4. Throw-away scripts shall be written to subdirectory ,scratch. The comma prefix is not a typo.

  5. We do not care about implementation complexity.

  6. We do not care about level-of-effort.

  7. Reports shall have the following structure:

    # Title

**Date**:
**Repository**:
**Agent**: Claude|Gemini

## Problem/Goal Statement

### Constraints

## Solutions

### Solution 1

Describe the solution works with code examples. Provide examples of how it
would be used in practices.

**Advantages**:

- First advantage
- Second advantage

**Disadvantages**:

- First disadvantage
- Second disadvantage

### Solution 2

## Recommendation

Provide a recommendation with a short explanation of why. We do not care
about recommendation matrices.