Gochal (고찰)
English | 한국어
A Claude Code plugin that transforms Claude from a solution-provider into a senior engineer thinking partner. Instead of listing A/B/C options and recommending one, Gochal (고찰) maps the solution landscape together with you — teaching trade-offs so you can make grounded decisions yourself.
Motivation
AI coding assistants tend to recommend solutions. The problem? Research shows that when AI presents recommendations, users follow them even when they conflict with their own judgment (Klingbeil et al., 2024). This short-circuits learning and produces decisions without understanding.
Gochal takes a different approach. Inspired by Applied Cognitive Task Analysis (ACTA) and preference construction theory, it:
- Teaches trade-offs instead of recommending solutions
- Simulates concrete scenarios before abstract discussion
- Lets you converge when you're ready — never pushes you toward a choice
- Produces transferable mental models — patterns you'll recognize in future decisions
The name 고찰 (go-chal) means "deep contemplation" in Korean. It reflects the skill's philosophy: thinking deeply together is more valuable than getting a quick answer.
How It Works
Gochal follows a gate + 5-phase cognitive protocol:
0. Gate — Decide whether full exploration is needed
1. Intent — Understand what you're building and why
2. Simulate — Walk through concrete end-to-end scenarios
3. Explore — Map the solution landscape for each decision point
4. Anchor — Surface core trade-offs as value questions
5. Converge — Confirm your decision with full context
→ Decision Record (고찰록)
Phase 0: Gate
Not every request needs full 5-phase exploration. Before entering, Gochal applies a "materially different" test: would the user reject one valid implementation in favor of another? If there's only one reasonable path, no unresolved high-risk constraints, and the request is specific enough — Gochal confirms and moves on without ceremony.
When context is missing, Gochal searches the codebase first instead of asking. It only asks when the artifact is genuinely undiscoverable.
Phase 1: Intent
Gochal reads your project context (files, docs, recent commits) and confirms understanding. Only asks what it genuinely doesn't know — no checklist interrogations.
When multiple unknowns exist simultaneously, Gochal resolves them in priority order: safety constraints first (production, data, migrations), then missing context (files, artifacts), then design branches. Safety answers often collapse the branch space entirely.
Phase 2: Simulate
Before any abstract discussion, you walk through 1-2 real scenarios step by step (3+ concrete steps each, not a one-sentence summary):
"Let's walk through the actual usage. A user arrives at X — what happens next?"
Each step surfaces constraints, assumptions, and decision points. Then hard cases are probed from two angles — technical failure ("what breaks?") and experiential failure ("what feels wrong even if it works?"). These scenarios become the evaluation criteria for Phase 3.
Phase 3: Explore
For each decision point, Gochal maps the landscape — not as a numbered menu, but as a guided exploration with enforced depth:
- Why each approach exists — specific problem context, not one-sentence labels
- Where each approach shines and breaks — at least 1 real-world example or failure case per approach
- Multiple perspectives — at least 2 distinct viewpoints per decision point ("performance engineer" / "product" / "maintainability")
- Mandatory scenario application — every approach is tested against Phase 2's hard cases
- Hybrid approaches — mixing elements from different approaches is encouraged
Phase 4: Anchor
Core trade-offs are framed as spectrum questions, not binary choices. Most design decisions have at least 2 independent trade-off axes — if only 1 is visible, the skill pauses to reconsider:
"It comes down to two axes: deployment complexity vs library access. Where do you lean on each?"
Phase 5: Converge
Only when you signal readiness ("let's go with this"). Gochal reflects the full picture — what you gain, what you trade off, and what patterns transfer to future decisions.
A Decision Record (고찰록) is generated — a shareable document capturing the landscape explored, trade-offs considered, and reasoning behind the decision.
Installation
# Add the marketplace
/plugin marketplace add yungjurick/gochal-plugin
# Install the plugin
/plugin install gochal-plugin@gochal-plugin
Usage
Invoke with any of these triggers:
/gochal-plugin:gochal
Or let Claude detect it automatically from natural language:
