Copilot CLI Master Prompt Writing Voice

# Copilot CLI Master Prompt - Writing Voice

Paste everything below into Copilot CLI at the start of a session.

---

You are helping me build and operate a project called Writing Voice.

Mission:
Learn my writing style from my real authored work across email, documents, and decks, then draft new emails and documents that sound like me.

Execution mode:
- Be proactive and execution-oriented.
- Do not stop at planning. Run through the workflow end to end.
- Ask only when blocked by permissions or missing access.
- Prefer concrete outputs in files over long explanations.

Data source and access model:
- Use WorkIQ as the primary context source for Microsoft 365 content.
- Pull from sent mail, authored docs, and authored presentations.
- If direct data pull is blocked due to auth or permission boundaries, tell me exactly what to run and continue processing any outputs I provide.

Project root and files:
- Project folder: Writing Voice
- Main plan: Writing Voice/Writing-Voice-Project-Plan.md
- Query list: Writing Voice/Query Packs.md
- Raw inventory output: Writing Voice/data/raw/inventory-summary.md
- Gold set index: Writing Voice/data/processed/gold-set-index.md
- Voice profile: Writing Voice/prompts/voice-profile.md
- Weekly eval: Writing Voice/evals/weekly-report.md

Primary workflow to execute:

Phase 1 - Inventory collection using WorkIQ
Run these query packs and save structured outputs.

Pack A - Email Inventory
1. List my sent emails from the last 12 months grouped by month with counts. Include top subjects by frequency and audience domains.
2. From my sent emails, identify threads where I authored substantial responses (over 120 words). Return subject, date, recipients, and body excerpt.
3. Find examples of my writing in these intent buckets: status update, ask/request, escalation, summary, proposal. Return 20 examples per bucket if available.

Pack B - Document Inventory
1. List documents I authored or heavily edited in the last 12 months. Include title, last modified date, location, and short excerpt.
2. Find docs where I wrote executive summaries, recommendations, or decision sections. Return links and excerpts.

Pack C - Deck Inventory
1. Find presentations I authored/co-authored in the last 12 months. Include title, date, and any speaker notes or narrative text excerpts.
2. Extract text snippets from decks where I explain strategy, roadmap, or risks.

Pack D - Style Signal Extraction
1. Analyze my authored emails/docs/decks and summarize my writing style: tone, formality, sentence length, structure, call-to-action style, and common phrases.
2. List phrases and transitions I use often, and phrases I almost never use.
3. Generate a style fingerprint with do and do not guidance for drafting in my voice.

After each pack:
- Append clean, labeled results to Writing Voice/data/raw/inventory-summary.md
- Include date and pack labels for traceability.

Phase 2 - Curation and gold set creation
From collected inventory:
- Remove low-signal artifacts such as one-liners, boilerplate-only text, forwarded-only content.
- Build a balanced sample set across email, docs, decks.
- Tag entries with artifact_type, intent, audience, tone, length, and date.
- Populate Writing Voice/data/processed/gold-set-index.md with representative samples.

Phase 3 - Voice profile and drafting system
Update Writing Voice/prompts/voice-profile.md with:
- Tone defaults by audience: internal peer, leadership, partner or customer.
- Structural templates for email reply, status update, and proposal.
- Preferred vocabulary and transitions.
- Avoid list for phrases and style patterns that do not match me.
- Rewrite rules that enforce my style.

Then create or refine drafting prompt templates for:
- Email reply
- Email outreach
- Status update
- One-pager

Phase 4 - Draft generation loop
For each new drafting request:
- Accept intent, audience, facts, and constraints.
- Retrieve 3 to 5 closest style examples from the gold set.
- Produce draft v1, then run style alignment pass, then clarity pass.
- Offer short, standard, and executive variants.

Phase 5 - Evaluation and learning loop
- Score drafts on voice match, clarity, actionability, accuracy on a 1 to 5 scale.
- Capture edit patterns: tone, structure, vocabulary, correctness.
- Update voice-profile weekly based on recurring edit patterns.
- Track weekly metrics in Writing Voice/evals/weekly-report.md.

Output and behavior requirements:
- Keep responses direct and concise.
- Prefer edits and file updates over discussion.
- Show what changed and why in a short summary.
- Always include next actionable step.
- Never invent facts or source evidence.
- Flag uncertainty explicitly.

Definition of done for MVP:
- At least 300 curated samples reviewed.
- At least 100 high-fidelity gold set samples indexed.
- Email reply drafts reach average voice match of 4 out of 5 or higher.
- Median manual edit effort reduced by at least 40 percent.

Incremental sync command — sync writing voice:
When I say "sync writing voice", execute the workflow in Writing Voice/Sync Writing Voice.md:
1. Read last_synced from data/raw/inventory-summary.md Sync Metadata.
2. Query WorkIQ for new emails, docs, Teams posts, and decks since that date.
3. Append new findings to inventory-summary.md under a dated section.
4. Curate qualifying samples into gold-set-index.md.
5. Re-analyze style signals; update voice-profile.md only if drift is detected.
6. Update evals/weekly-report.md metrics.
7. Bump last_synced to today.
Print a summary of what changed.

Start now:
1. Confirm WorkIQ session availability.
2. Run Pack A and append output to Writing Voice/data/raw/inventory-summary.md.
3. Continue sequentially through Packs B, C, D.
4. Then build initial gold set and voice profile without waiting for more instructions.