| name | brainstorming-prompt-file |
| description | Use this before spawning coding agents when you need to turn an idea into an execution-ready prompt file for bootstrap-task.sh or spawn-agent.sh. |
Brainstorming Ideas Into Agent Prompt Files
Overview
Turn a rough idea into a concrete prompt file that a coding agent can execute with minimal ambiguity.
This skill is the prompt-file-oriented variant of the Socrates brainstorming flow, optimized for orchestration workflows where output must feed directly into:
--prompt-file for bootstrap-task.sh or spawn-agent.sh
--completion-mode pr|no-pr-spec
- optional no-PR spec file checks
The Process
1) Understand The Objective
- Start by checking local project context first (files, docs, recent changes)
- Ask one question at a time
- Prefer multiple choice when possible
- Clarify:
- target repo
- desired outcome
- constraints
- completion mode (
pr or no-pr-spec)
2) Explore Approaches
- Propose 2-3 approaches with trade-offs
- Lead with your recommendation and why
- Keep suggestions practical and YAGNI-focused
3) Build The Prompt In Sections
Draft the prompt file incrementally in short sections and validate each section with the user:
- Goal and business context
- Scope and non-goals
- Technical constraints and touched areas
- Implementation requirements
- Validation steps and done criteria
If anything is unclear, go back and ask one follow-up question.
4) Finalize Execution Inputs
Produce:
- Prompt file content
- Suggested file path (default:
tasks/prompts/YYYY-MM-DD-<topic>.md)
- Ready-to-run command using
bootstrap-task.sh
If completion-mode=no-pr-spec, also ensure there is a spec file and reference it in the command.
Use scripts/init-spec.sh when needed.
Prompt File Contract
Every generated prompt file should include:
- Objective
- Why this matters (brief business context)
- In scope
- Out of scope
- Implementation notes (files/components/services likely involved)
- Constraints (performance, compatibility, security, no schema changes, etc.)
- Validation (commands/tests/checks to run)
- Definition of done
- Output format expected from agent (summary, files changed, risks)
Completion-Mode Rules
pr mode
Prompt should instruct the agent to:
- commit and push branch
- open/update PR
- address CI and review comments
no-pr-spec mode
Prompt should instruct the agent to:
- produce required artifacts
- satisfy spec checks in the referenced spec file
- report evidence for each check
Output Behavior
When the user says the prompt is approved:
- Write the prompt file to disk
- Print the exact
bootstrap-task.sh command to run
- Ask if they want to spawn immediately
Key Principles
- One question per message
- Prefer multiple choice
- Remove unnecessary complexity
- Make success criteria explicit
- Optimize for execution quality, not essay quality