Product Analysis

---
description: "Deep product analysis via Ralph loop   feature gaps, UX issues, tech debt, innovations   powered by 16 CLI tools"
allowed-tools: Read, Write, Edit, MultiEdit, Bash, Glob, Grep, Agent, WebFetch, WebSearch, AskUserQuestion, Skill, EnterPlanMode, ExitPlanMode, TaskCreate, TaskGet, TaskUpdate, TaskList, TodoWrite, NotebookEdit, ToolSearch, mcp__desktop-commander__*, mcp__chrome-devtools__*, mcp__claude-in-chrome__*, mcp__memory__*, mcp__context7__*, mcp__clear-thought__*, mcp__playwright__*
argument-hint: "[--iterations N] [--skip-interactive]"
---
# Product Analysis Command
Deep 5-phase product analysis using the Ralph Wiggum loop for iterative refinement. Works on ANY project   any language, framework, or domain. Produces two deliverables:
- `docs/product-analysis-report.md`   comprehensive report with quantitative evidence
- `docs/product-analysis-checklist.md`   actionable items with IDs, effort, and priority

## Skill Location
All templates, scripts, and references live in:
`~/.claude/skills/product-analysis/`
- `SKILL.md`                          Router + overview (read for phase details)
- `assets/report-template.md`         Ralph loop prompt template (with {{PLACEHOLDERS}})
- `references/`
  - `domain-detection.md`           Domain   competitor   persona mappings
  - `cross-phase-analysis.md`       Cross-referencing strategies for iteration 6+
  - `examples.md`                   Sample reports for different project types
- `scripts/`
  - `detect-project.sh`             Auto-detect project context   JSON
  - `fill-template.sh`              Replace {{PLACEHOLDERS}}   filled prompt
  - `baseline-metrics.sh`           Run scc/tokei/lizard/git   metrics table
  - `validate-report.sh`            Verify deliverables pass quality gates

## Step 1   Parse Arguments
Check `$ARGUMENTS` for flags:
- `--iterations N`   use N as loop count (each loop = 5 iterations), skip interactive question
- `--skip-interactive`   use default of 2 loops (10 iterations)
- No arguments   proceed to interactive question

## Step 2   Ask Iteration Count (unless skipped)
If no `--iterations` flag was provided and `--skip-interactive` was not set, ask the user:
Use `AskUserQuestion` with this exact configuration:
- **Question**: "How many full analysis loops? Each loop = 5 phases (inventory, gaps, UX, debt, innovation). More loops = deeper refinement."
- **Header**: "Depth"
- **Options**:
  1. Label: "1 loop (5 iterations)"
     Description: "Quick scan   covers all phases once"
  2. Label: "2 loops (10 iterations) (Recommended)"
     Description: "Standard depth   covers all phases with one refinement pass"
  3. Label: "3 loops (15 iterations)"
     Description: "Deep analysis   thorough multi-pass refinement"
  4. Label: "Custom"
     Description: "Up to 10 loops (50 iterations max)"

If user selects "Custom", ask a follow-up:
- **Question**: "How many loops? (1-10, each loop = 5 iterations)"
- **Header**: "Loops"
- **Options**:
  1. Label: "5 loops (25 iterations)"
     Description: "Extended analysis"
  2. Label: "7 loops (35 iterations)"
     Description: "Comprehensive deep dive"
  3. Label: "10 loops (50 iterations)"
     Description: "Maximum depth   exhaustive analysis"

Compute: `max_iterations = loops * 5`

## Step 3   Auto-Detect Project Context
Run the detection script to gather project information automatically:
```bash
bash ~/.claude/skills/product-analysis/scripts/detect-project.sh "$(pwd)"

This outputs a JSON object with: project_name, tech_stack, domain, project_root, key_files, competitors, personas, date, baseline. Save the output to a temporary file:

bash ~/.claude/skills/product-analysis/scripts/detect-project.sh "$(pwd)" > /tmp/product-analysis-context.json

If the script fails (missing jq, scc, etc.), fall back to manual detection:

1. Read CLAUDE.md or README.md for project name and description
2. Read package.json / Cargo.toml / pyproject.toml / go.mod for stack
3. Infer domain from dependencies (read ~/.claude/skills/product-analysis/references/domain-detection.md for signal tables)
4. Manually construct the JSON object and write it to /tmp/product-analysis-context.json

Review the detection output. If the project name or domain looks wrong, use your judgment to correct it before proceeding. The script reads package.json name field which may differ from the actual project name check CLAUDE.md or the directory name.

Step 4 Fill Template and Generate Prompt

Use the fill-template script to replace all {{PLACEHOLDERS}} in the template:

bash ~/.claude/skills/product-analysis/scripts/fill-template.sh   ~/.claude/skills/product-analysis/assets/report-template.md   /tmp/product-analysis-context.json   .claude/ralph-product-analysis.md

This reads assets/report-template.md, replaces {{PROJECT_NAME}}, {{TECH_STACK}}, {{DOMAIN}}, {{PROJECT_ROOT}}, {{KEY_FILES}}, {{COMPETITORS}}, {{PERSONAS}}, and {{DATE}} with values from the context JSON, and writes the filled prompt to .claude/ralph-product-analysis.md (project-local).

If the script fails, read ~/.claude/skills/product-analysis/assets/report-template.md and manually replace placeholders using the context JSON values. Write the result to .claude/ralph-product-analysis.md.

Verify: Check that .claude/ralph-product-analysis.md exists and contains no {{PLACEHOLDER}} text (except in example/comment sections):

grep -c '{{[A-Z_]*}}' .claude/ralph-product-analysis.md

This should return 0 or a very small number (only from example text within the template).

Step 5 Display Summary

Before launching the loop, display a summary to the user:

Product Analysis Configuration
==============================
Project:      [project_name from context]
Stack:        [tech_stack from context]
Domain:       [domain from context]
Competitors:  [competitors from context]
Personas:     [personas from context]
Iteration