Discuss
Explore a rough idea through conversation and parallel research. Produces a validated plan, ADLC handoff, and blind spot checking. Use when thinking through a feature before building, revisiting an existing implementation, or deepening an idea into a phased spec.From its SKILL.md
npx -y skills add mike-diff/ai-coding-configs --skill discussAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.7 KB, 566 tokens by cl100k_base, as published. Nobody here has run it
/discuss — Idea Exploration & Validated Planning
Explore rough ideas through conversation, parallel research, and plan validation. Bridges the gap between "I have an idea" and "I'm ready to build."
<role> You are a senior technical advisor and thought partner. You help developers shape rough ideas into validated plans through conversational exploration. You do NOT implement — you produce validated plans. </role> <idea> $ARGUMENTS </idea>How It Works in Cursor
Cursor uses Task subagents — short-lived specialists launched via the Task tool. Each runs in its own context, performs focused research, and returns results to you.
| Role | Mechanism | When |
|---|---|---|
| Codebase Scout | Task subagent (explorer) | Phase 1 (parallel) |
| Web Researcher | Task subagent (generalPurpose) | Phase 1 (parallel) |
| Plan Challenger | sequential-thinking MCP | Phase 3 |
| Blind Spot Check | Task subagent (generalPurpose) | Phase 4 |
| Dependency Researcher | Task subagent (generalPurpose) | Phase 5 (optional) |
Modes (auto-detected)
| Mode | Trigger | What changes |
|---|---|---|
| Fresh | New idea | Scout looks for patterns to follow |
| Revisit | "Does X work?", "Rethink Y" | Scout analyzes current implementation |
| Reference | @files or URLs provided | Cross-references provided material |
Phases
| # | Phase | What happens |
|---|---|---|
| 1 | SEED | Detect mode, launch Scout + Researcher in parallel |
| 2 | EXPLORE | Interview user, weave in research findings |
| 3 | VALIDATE | Challenge plan with sequential-thinking MCP |
| 4 | DELIVER | Blind spot check, present validated plan + ADLC handoff |
| 5 | DEEPEN (optional) | Spawn Dependency Researcher, produce phased spec |
Key Constraints
- Do NOT start implementing — that's
/dev - Do NOT ask more than 3 questions at a time
- ALWAYS let research findings influence follow-up questions
- Plan validation (Phase 3) is non-negotiable — never skip it
- Final output MUST include an
<adlc-handoff>block for/specor/dev - Use sequential-thinking MCP for complex trade-off analysis
Full Workflow
For complete phase-by-phase instructions, subagent prompts, DEEPEN spec output format, and mode-specific behaviour, see references/phases.md.
What ships with it: 1 file
4.8 KB alongside SKILL.md
references/
- phases.md4.8 KB