dignified-technology

The core question: does this tool owe its users their own capability back, amplified — or does it take something from them in exchange?

This is a diagnosis-first hybrid approach. The skill gathers as much evidence as possible from code before asking any questions, making the interview shorter and more focused.

<required_reading> Read these reference files before proceeding:

1. references/essay-principles.md — The 6 principles with full context
2. references/scoring-rubric.md — A-F grade definitions per principle
3. references/values-as-spec-guide.md — Values-as-Spec assessment guide
4. templates/scorecard.md — Output template for the final report </required_reading>

<quick_start> Run /dignified-technology in any project directory. The skill will:

1. Diagnose — Read your codebase (CLAUDE.md, README, source code, docs) to form preliminary assessments
2. Interview — Ask targeted questions only about gaps the diagnosis couldn't resolve
3. Assess values — Ask three Values-as-Spec questions (always asked, requires human intent)
4. Score — Produce a letter-grade scorecard with per-principle grades and recommendations </quick_start>

Read the following files and sources (skip any that don't exist):

1. CLAUDE.md — Product description, architecture, design philosophy
2. README.md — Product overview, stated purpose, user-facing description
3. PRDs, design docs, or values statements — Search for files matching: **/PRD*, **/prd*, **/design-doc*, **/values*, **/principles*, **/ARCHITECTURE*
4. AI integration points — Search source code for:
   • AI/LLM API calls (anthropic, openai, ai, llm, completion, chat, generate, prompt)
   • Streaming endpoints (SSE, stream, EventSource)
   • AI-related route handlers
5. User-facing flows — Examine UI components, pages, and user interactions to understand:
   • How AI output is presented to users
   • What controls users have over AI behavior
   • Whether the user is involved in the creation process or receives finished output
6. Notification and suggestion patterns — Search for unsolicited AI behavior, auto-complete, auto-suggest, notifications

For each of the 6 principles, record:

• Tentative grade (based on code evidence alone)
• Evidence found (specific files, patterns, design decisions)
• Gaps (what you couldn't determine from code — these become interview questions)

Present the diagnosis to the user as a summary table:

Based on reading your codebase, here's my preliminary assessment:

| Principle | Tentative Grade | Evidence | Gaps |
|-----------|----------------|----------|------|
| Authorship Preservation | [grade] | [what was found] | [what's unclear] |
| Creative Range | [grade] | [what was found] | [what's unclear] |
| Voice Amplification | [grade] | [what was found] | [what's unclear] |
| Process Involvement | [grade] | [what was found] | [what's unclear] |
| Depth of Exploration | [grade] | [what was found] | [what's unclear] |
| Input Agency | [grade] | [what was found] | [what's unclear] |

I need to ask you some questions to fill the gaps before finalizing grades.

Rules for this phase:

• Only ask about gaps from Phase 1 — if the code clearly answers a question, don't re-ask it
• Frame questions based on what was found: "I saw that your AI generates drafts in routes/writing.ts. Can you tell me about the intent behind that design? Is the user expected to rewrite the draft substantially, or is it meant as near-final output?"
• Group related questions when possible to reduce interview length
• Maximum 6 questions total (one per principle gap, though some principles may have no gaps)