iorlas-brainstorm

Adaptive brainstorming & decision coaching for Claude. Shift-left your decisions.

An AI skill that detects your decision-making pattern — rationalizing, over-analyzing, avoiding commitment, navigating politics, or genuinely torn — then applies targeted techniques to surface the real issue and produce actionable outcomes.

Quick Install (via Marketplace)

claude plugin marketplace add iorlas/iorlas-marketplace
claude plugin install iorlas-brainstorm@iorlas-marketplace

Manual Install (Claude Code)

Copy the skill file directly:

mkdir -p ~/.claude/skills/iorlas-brainstorm
cp skills/iorlas-brainstorm/SKILL.md ~/.claude/skills/iorlas-brainstorm/SKILL.md

Or clone the repo and install:

git clone https://github.com/iorlas/iorlas-brainstorm.git
mkdir -p ~/.claude/skills/iorlas-brainstorm
cp iorlas-brainstorm/skills/iorlas-brainstorm/SKILL.md ~/.claude/skills/iorlas-brainstorm/SKILL.md

Claude.ai / Claude Desktop

1. Download or clone this repo
2. Zip the skills/iorlas-brainstorm/ folder
3. Upload the zip as a project file in Claude.ai or Claude Desktop

What to Expect

When you invoke /iorlas-brainstorm, the skill runs a 5-phase adaptive session:

1. Tool Fit Assessment — Checks whether brainstorming is the right tool (vs. emotional processing, crisis support, or data gathering)
2. Situation Assessment — Gathers context with focused questions, detects your communication style
3. Mode Classification — Silently classifies your decision-making pattern into one of 5 modes:
   • Adversarial — You're rationalizing a pre-made decision
   • Validated Design — You've done the work, you need execution help
   • Commitment Push — You already know the answer but can't commit
   • Political Navigation — The real problem is organizational, not technical
   • Ambiguity Framework — The trade-offs are genuinely balanced
4. Mode-Specific Engine — Applies targeted techniques for your pattern
5. Convergence — Produces a structured take-away document with concrete action items, behavioral scripts, and self-limitation acknowledgments

Micro-Commands

At any point during a session, you can say:

• steelman — Steelman the opposing position
• premortem — Run a premortem (past tense, 3-4 failure modes)
• invert — What would guarantee failure?
• blind spots — Find blind spots
• advisor: [name] — Get a named expert's perspective
• red team — Red team exercise
• decide — Jump to convergence
• mode: [name] — Force a mode switch
• map decisions — Map multiple interconnected decisions

Variants

Variant	File	Size	Avg Score	Notes
Compact (default)	SKILL.md	7,675 chars	90.9%	Compressed prompt — higher benchmark scores, lower token cost
Verbose	SKILL-verbose.md	30,982 chars	84.6%	Original uncompressed prompt — full explanations and examples

Both variants implement the same 5-phase session, the same 5 modes, and the same micro-commands. The compact version was produced through prompt compression research and validated across 48 simulations.

Evaluation

Tested against 5 other brainstorming skills across 8 scenarios (48 total simulations) using an 11-dimension rubric (max 100/scenario):

Rank	Skill	Avg Score	σ
1	iorlas-brainstorm v5-compact	90.9%	2.9%
2	v5-modular	90.1%	3.5%
3	v4	88.9%	5.9%
4	BMAD	87.9%	3.5%
5	superpowers	86.1%	2.5%
6	v5-original (verbose)	84.6%	2.6%

Methodology note: Each data point is a single self-play simulation (N=1 per cell, overall σ=3.8pp). Gaps below ~5pp are within noise. All simulations used Claude as the backbone model. See evaluations/ for full results, rubric, persona scripts, and re-run instructions.

Platform Compatibility

Platform	Method	Status
Claude Code (CLI)	Plugin install or manual ~/.claude/skills/	Supported
Claude Desktop	Project file upload (zip)	Supported
claude.ai	Project file upload (zip)	Supported
Cursor / Copilot / OpenCode	Place SKILL.md per platform conventions	Compatible

License

MIT