jahidulislamseo/antigravity-ai-skills

Brief Management

Briefs are persistent work state for local SEO engagements. One brief per location. Brand rollup for multi-location work. Scheduled task outputs extend the brief over time, building a compounding history of every scan, audit, report, and finding for each business.

---

Structure

briefs/
  {brand}/
    _brand.brief.md          ← config + rollup across all locations
    reports/                 ← brand-level rollup reports
    {location}/
      location.brief.md      ← always current, always lean
      reports/               ← weekly, monthly, QBR reports
      scans/                 ← geogrid scans, citation audits, page audits
      drafts/                ← GBP posts, review responses awaiting approval
      alerts/                ← monitoring alerts

---

First Run Detection

This is the most important section. When a user mentions a specific business and no brief exists, do not proceed with their request yet. Run setup first, then circle back to what they asked for.

Detection

Before doing any work for a specific business, check for an existing brief:

• Claude Code: look for briefs/{brand-slug}/{location-slug}/location.brief.md
• Claude Project: check Project knowledge base for a brief file for this business
• If found: go to Resuming from a Brief
• If not found: go to First Run Setup

First Run Setup

Tell the user:

"I don't have a brief for [Business Name] yet. Let me grab a few details so I can track everything going forward. This takes about 2 minutes."

Then ask these questions one at a time, not all at once:

Q1: "What's the primary keyword you want to rank for? (e.g. 'plumber buffalo ny')"

Q2: "What's the website URL?"

Q3: "How far does the service area extend? (e.g. 5 miles, 20 miles)"

Q4: "Are you managing this for a client, or is this your own business?"

• If client, ask: "What's the client's email for reports?"
• If own business, skip

Q5: "Do you want alerts and reports sent to a Slack channel?" (optional, skip if they say no)

That's it. Don't ask for more than these unless something is genuinely ambiguous.

After Questions

1. Create brief folder structure:

   • briefs/{brand-slug}/{location-slug}/location.brief.md
   • briefs/{brand-slug}/{location-slug}/reports/
   • briefs/{brand-slug}/{location-slug}/scans/
   • briefs/{brand-slug}/{location-slug}/drafts/
   • briefs/{brand-slug}/{location-slug}/alerts/
   • briefs/{brand-slug}/_brand.brief.md

2. Populate Identity section from answers and any context already in the conversation

3. Configure _brand.brief.md approval block from Q4/Q5 answers

4. Run initial audit automatically, don't ask permission:

   • LocalSEOData local_audit
   • LocalSEOData business_profile + profile_health
   • LocalSEOData geogrid_scan (7x7, primary keyword, stated radius)
   • LocalSEOData citation_audit
   • LocalSEOData review_velocity + google_reviews
   • LocalSEOData competitor_gap

5. Populate Findings from audit results

6. Set Next Action to the single highest-priority fix

7. Offer scheduled task setup:

   "Brief created and initial audit done. I found [X critical, Y important issues]. [Top finding in one sentence].

   Want me to set up automated monitoring? I can configure weekly ranking scans, review monitoring, GBP change detection, and monthly reports to run automatically."

   If yes, go through task list from tasks/README.md, recommend based on audit findings, configure via /schedule If no, proceed with original request

8. Then return to what the user originally asked for:

   "Now let me [original request]..."

For Claude Project / Browser Users

If no filesystem access: run the same conversation and audit, then output the brief as formatted text at the end of setup:

"Here's your brief. Add this to your Project knowledge base so I'll have it next time:" [brief content]

---

When to Create a Brief (Non-First-Run)

Outside of first run, create a brief mid-session if:

• A tool call is about to be made for a business with no brief
• User explicitly asks to create one

Do not create briefs for:

• General local SEO questions with no specific business
• Tool or skill explanations
• Single-question answers with no live data

---

Brief Location

Claude Code / Cowork:

briefs/{brand-slug}/{location-slug}/location.brief.md

Claude Project: Upload brief file to Project knowledge base after creation. Tell user to do this if they haven't.

---

Updating a Brief (Manual Sessions)

At end of every session:

1. Session Log. One line: date + what was worked on
2. Tools Run. Each tool call: date, tool, endpoint, one-line finding
3. Findings. New issues, categorized Critical / Important / Monitor
4. Deliverables. Mark completed, add new pending
5. Next Action. Overwrite with single most important next step

---

Receiving Scheduled Task Output

When a scheduled task writes an output file:

1. Read the output file
2. Add one line to Session Log: [DATE] | {task type} → {one-line finding} → see {output path}
3. If new Critical findings: update Findings section
4. Update Next Action if the task surfaced a higher priority than current
5. Update _brand.brief.md Locations table row for this location

The brief stays lean. Detail lives in the output files. The brief is the index.

---

Resuming from a Brief

When a user returns to a business:

1. Read the brief
2. Read the most recent output file in each subfolder for current state
3. Summarize in 2-3 sentences: what's been done, key findings, current state
4. State the Next Action
5. Ask: "Want to pick up from here, or something specific?"

Do not re-run tools with results in scans/ unless user asks or data is >30 days old.

---

The Compounding Value

After months of scheduled tasks writing to a brief, the history is queryable:

• "Why did rankings drop in March?" reads scans/ for that period
• "How have reviews trended this year?" reads reports/ review monitors
• "When did this competitor appear?" reads competitor monitors

Never delete output files. Archive if the folder gets unwieldy. The history is the value.

---

Lessons Flywheel

During every session, manual or scheduled, watch for:

• Tool returning data that contradicts a skill file
• A pattern emerging across 3+ locations not documented anywhere
• GBP, algo, or AI platform behavior differing from documented expectations
• A tool endpoint behaving unexpectedly

When observed, surface it explicitly:

"I noticed [X] during this session. This pattern isn't in the skill files. Want me to add it to meta/lessons.md for review?"

Wait for confirmation. Then write the lesson.

---

Multi-Location Brands

After each location session:

• Update that location's row in _brand.brief.md Locations table
• Add to Brand-Level Notes only if pattern appears in 3+ locations

---

Slug Format

Lowercase, hyphens, obvious:

• Brand: keystone-insurance, mikes-plumbing
• Location: buffalo, downtown-chicago, pittsburgh-pa