What this session did

Rebuilt the copywriting skill from generic SaaS boilerplate into the firm's PI web-copy engine and callout command. Two modes: Mode A net-new copy, Mode B full-scale stale-to-new revision of an existing page or post, invokable as /copywriting <url> or /copywriting new <brief>. Branches by page type (location, practice-area, post, general).

Captured 31 rules from you in rapid fire and wrote them in, self-contained: header casing matrix (H1/H2 all caps mandatory, H3-H5 optional, no H6), centered headers/titles, no eyebrows or pills, no compound headers, no header starting with "The", no "Why It Matters", em-dash ban with a spaced-hyphen allowed only for attribution, one UVP per page, no negative parallelisms, plain headers, split-card hero family (no single full-width image block, no default trust strip, uniform overlay), full-width blocks + desktop inspection + leaky-CSS autofix, FAQ bottom-anchored, On This Page dropdown, device-aware CTA (desktop GET STARTED / mobile CALL NOW), CTA centered below text, sidebar container spacing, no two-column odd-numbered lists on desktop, review and results verification gate, ALL-CAPS review names (first + last initial primary, first initial + last name fallback), no full names, no "Google verified" text, real Google symbol in approved layout, featured image required + visual inspection + AI-slop handling, relevant-video proposal, QA during the revision, Playwright dark-background contrast check.

Folded the canonical banned-words list (sourced from brand-voice-enforcer) and the AI-tell inventory directly into the skill plus two bundled reference files. Built, validated with skill-creator ("Skill is valid"), and packaged copywriting.skill. Tested the engine on a live page (louisville-sideswipe-accidents) and produced a clean, rule-compliant rewrite as proof.

Decisions locked this session

Banned words and AI tells inline (fully self-contained). Deploy autonomy: auto-deploy non-home pages with rollback, home page never without permission. Page-type and post awareness on. Slash-command invocation. 5A: make copywriting the canonical web-content engine and repoint dead references.

Problems and things caught

• Dangling references: wp-content-deployer, pre-publish-qa, and playwright-visual-qa all point to web-content-new / web-content-revise / location-pages / practice-area-content skills that do not exist in the library. copywriting now fills that role. Repointing is prepared but not applied (installed skills cannot be edited from the web interface).
• Cross-skill drift: brand-voice-enforcer says primary conversion is phone calls, which the new device-aware desktop GET STARTED CTA refines. Needs a sync.
• The approved tagline "Hurt in a car? Call Sam Aguiar!" brushes against the new no-hurt-in-a-wreck-UVP rule. Your call on whether the established tagline is exempt.
• I initially proposed referencing ai-writing-detection instead of writing the tells in; you corrected it to self-contained. Fixed.

Why the session ended

Reached the build, package, and test milestone. Live deployment, slash-command installation, and the 5A repointing need the connected environment (Cowork or Claude Code); flagged for follow-up rather than faked from the web sandbox.

Recommended next actions (suggestions, your discretion)

1. Install copywriting.skill and confirm the slash command runs.
2. In Cowork, apply the 5A repointing to the three dependent skills (one line each).
3. Run Mode B live on a low-stakes page as a draft first, eyeball it, then turn on autonomous-with-rollback.
4. Sync brand-voice-enforcer to the device-aware CTA; confirm the tagline exemption.
5. Optional: build the cloud-resident endpoint (Cloudflare Worker or hosted MCP) for any-agent Bearer access.
6. Confirm the hero overlay opacity value and the title-case function-word handling (both minor).