Copy Command

Before you run this skill

This skill is brand-neutral. It reads its brand, palette and endpoints from brand.config.json at the repo root.

On first use, do this before anything else:

1. Run python3 brandkit.py. It prints the config source and any placeholder that is still unset.
2. If it says configured: False, copy brand.config.example.json to brand.config.json.
3. Ask the operator for each value under missing, then write them in. Do not guess a brand name, a domain, or a colour.
4. Anything the skill writes out should be passed through brandkit.fill(text), which swaps every {{TOKEN}} for its configured value and remaps the default palette to the operator's.

Text below uses {{TOKEN}} where a value is operator-specific. Treat an unresolved {{TOKEN}} in your output as a bug, not as literal copy.

/copy-command — Email Copy → Google Sheet

Takes approved email copy from a Google Doc and places it into a Google Sheet, structured the way the user asks, with clean personalization and zero signature contamination. Prioritizes accuracy, clean personalization, correct segmentation, and campaign readiness over speed.

Step 0 — Load prior learnings (MANDATORY, before anything else)

Read LEARNINGS.md in this skill's directory (~/.claude/skills/copy-command/LEARNINGS.md). It is the accumulated record of corrections and confirmed behaviors from every previous copy-management iteration in this project. Everything in it is default behavior — apply it without the user having to repeat it.

At the END of every run: if the user corrected anything, or you established a new convention (a new banned phrase, a new mapping rule, a client-specific quirk), append it to LEARNINGS.md with the date and one line of context.

Step 1 — Collect inputs

Ask the user for all of these up front (one message / one AskUserQuestion batch, not a drip). Skip any the user already provided.

1. Sheet URL — the Google Sheet to add copy to.
2. Tab name — the tab (or base name) the copy goes into.
3. Tab structure — ask: "Do you want to create a new tab or multiple tabs? Describe how you want the tabs structured."
4. Subject line — ask: "Do you want to add a subject line? (y/N)" If yes, get the subject line(s) or confirm they come from the copy doc.
5. Copy Document URL — the Google Doc containing the approved email copy.
6. Multiple-copy clarification — ask: "If the document contains multiple copy variants, paste the copies here and explain which copy should be used for which tab/segment." If the doc has one variant, this can be skipped.

Do not start writing to the sheet until the copy→tab mapping is unambiguous. If the doc contains more variants than the user mapped, stop and ask.

Step 2 — Read the sources

• Copy doc: read via the Docs API through the OAuth helper (readonly scope is already on the token). Extract each variant exactly as written: subject (if present), email 1, and every follow-up in order.
• Sheet: read the source tab(s) via the Sheets API. Identify the lead columns (first name, company, email, any segment/ICP columns) and note every existing column — all of it must survive untouched.

Step 3 — Clean personalization variables (ALWAYS, via /copy-cleaner)

Before any copy lands in the sheet, clean the personalization inputs for every lead by invoking the copy-cleaner skill (~/.claude/skills/copy-cleaner/SKILL.md). It runs the LOCKED prompt v6 (operator-approved 2026-08-18, Prompt Builder task copy-personalization-cleaner) and is authoritative for all three fields:

• first_name_clean — the one name the person goes by (Muhammad Ali Kazimi → Ali, Dr Karim El Solh → Karim, Mariia Eroshin, JD, CFA → Mariia).
• company_clean — shortest form that still uniquely identifies the firm (Transpose Platform → Transpose, Raffles Family Office → Raffles, Partners Capital stays).
• event_clean — the 1-3 word name attendees call the event (SuperReturn Asia Venture Capital Summit 2026 → SuperReturn, Markets Group Private Equity Boston Forum 2026 → PE Forum) — for event-based campaigns (subject lines and opening lines per the 2026-08-18 event-copy learning).

Execution per that skill: dedupe unique values, fan out sub-agents that read the locked prompt file cold and verbatim, merge by row id, honor its statuses:

• unresolvable_name / unresolvable_company → null: blank the copy cell or use the fallback branch, flag in Notes, report the count. Never render "Hi NULL" or "your work at LLC".
• needs_review → the value ships, but list those rows in the run summary.

Do NOT re-implement cleaning with ad-hoc heuristics or a paraphrased prompt — drift was proven in tuning. Legacy references (CNCleanup, scripts/push_to_drive/ib_clean_first_names.py) are superseded by copy-cleaner for this skill's cleaning stage; a Python fast path is acceptable ONLY if its outputs match copy-cleaner's golden examples table exactly (spot-check before use).

Step 4 — Prepare the copy

• Preserve the original copy. Do not rewrite, reword, "improve", or shorten the approved copy. The ONLY permitted changes: personalization substitution, company-name cleaning, first-name cleaning, signature removal, "Sent from my iPhone" removal, and variant→tab mapping.
• NEVER add a signature. No Ramo, no Dash, no Best, [Name], no Thanks, [Name], no sender sign-off of any kind. If the source doc ends with one, strip it. The sending platform appends the signature dynamically.
• NEVER include "Sent from my iPhone" or any equivalent mobile signature. If the source doc contains it, remove it before writing to the sheet.
• Multiple variants: identify each variant clearly, keep its follow-ups in the correct number and order, and map it to exactly the tab/segment the user specified. Never mix lines from one variant into another. If the user asked for a randomized split, make it deterministic (sort-and-slice on a hash of a stable lead key, per the linkedin-outreach-sheet convention) so re-runs land the same leads in the same buckets.

Step 5 — Write to the sheet

Writes go ONLY via the Python OAuth helper ({{GOOGLE_ACCOUNT}}). Never use the Drive MCP create/copy tools — they are bound to a different account ({{BRAND_NAME}}).

import sys
sys.path.insert(0, "{{WORKSPACE_DIR}} - Second Brain")
from scripts.push_to_drive.auth import sheets_service, docs_service

Run scripts with the project venv:

cd "{{WORKSPACE_DIR}} - Second Brain/scripts/push_to_drive"
./.venv/bin/python <your_script.py>

Sheet rules (unless the user explicitly instructs otherwise):

• Never modify the original/source tab. New copy goes into new tabs or new columns per the user's requested structure.
• Preserve all existing lead data and columns. Never delete a column. Columns not needed by the VA/campaign may be hidden, not removed.
• New tabs follow the house format: Geist font everywhere, bold frozen header row, copy columns (Email Body, Followup 1, Followup 2, LinkedIn DM) at ~220px width, other columns auto-resized.
• Subject line goes in its own Subject Line column when requested.
• Follow-ups each get their own column, in send order.
• Write per-lead rendered copy (personalization substituted), not raw templates, unless the user asks for template + variable columns.
• If the token is expired, run reauth.py in the same directory first.

Step 6 — Final quality check (MANDATORY before reporting done)

Verify, by re-reading what was actually written to the sheet:

• Every intended lead has the correct copy variant.
• Personalization variables are populated for every row (no [First name] / {{company}} tokens left unrendered, no blanks mid-sentence).
• Company names are clean and natural (copy-cleaner output, spot-checked against its golden examples).
• First names are clean; event names (if used) are the copy-cleaner short form.
• No null/unresolvable lead got a rendered greeting or company mention; all flagged rows are counted in the report.
• Subject lines are present when requested, absent when not.
• Follow-ups are in the correct order, correct count per variant.
• Variants are correctly separated — spot-check rows in each tab against the mapping.
• No signature anywhere (Ramo, Dash, Best,, Thanks, …).
• No "Sent from my iPhone" or equivalent.
• No copy from one variant bleeding into another.
• Source tab untouched; no existing columns deleted.
• Tabs are campaign-ready as-is.

Report the sheet URL, which tabs were created/updated, lead counts per tab, and anything you cleaned or flagged (e.g. "14 first names fixed, 2 left blank and flagged in a Notes column").

Step 7 — Capture learnings

If this run produced any correction from the user or any new convention, append it to LEARNINGS.md (dated, one line of context). That is how this skill gets better instead of repeating mistakes.