First Principles

4. Explore Approaches

For each novel subproblem, identify 2-3 approaches:

• What's the simplest approach?
• What's the most robust approach?
• What would an expert do?

5. Evaluate Trade-offs

For each approach:

• Complexity (implementation + maintenance)
• Risk (what could go wrong)
• Reversibility (how hard to change later)

6. Decide

• Select an approach with clear reasoning
• Document what we're explicitly NOT doing and why
• Define the first concrete step

Output Mode

By default, output the analysis in chat. If the argument includes a file path (e.g., /first-principles "auth redesign" --output analysis.md), or if the problem is complex enough to warrant a persistent artifact, write the full analysis to a markdown file instead.

When writing to file:

1. Write first-principles-analysis.md (or user-specified path) in the project root
2. Include all 6 steps with findings, trade-offs, and the action plan
3. Tell the user: "Analysis written to <path> — review, annotate inline, and I'll iterate"
4. If the user re-invokes with the same file, read their inline annotations and update the analysis

This produces a persistent, reviewable artifact rather than ephemeral chat output.

Rules

• Use sequential-thinking MCP for structured reasoning when the problem has many dimensions
• Read relevant code before proposing approaches — ground recommendations in reality
• Challenge assumptions — ask "why" at least once for each stated constraint
• Prefer the simplest approach that meets requirements
• If the user's framing of the problem seems wrong, say so
• Output a clear action plan at the end, not just analysis