GeminiSAFE

Open-source SEO, GEO, and marketing skills for AI agents.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: gemini
argument-hint: "'review', 'challenge', or 'consult' + optional context"
description: >
  Cross-model second opinion from Google Gemini — a different AI reviewing the
  same changes, with deep Google ecosystem knowledge. Three modes: review
  (pass/fail gate for Google Ads campaigns, SEO metadata, or code), challenge
  (adversarial stress-test that tries to break your changes), and consult
  (open Q&A with Gemini on Google Ads strategy, SEO best practices, or
  implementation questions). Use when the user says "gemini review", "ask
  gemini", "gemini challenge", "second opinion from gemini", "consult gemini",
  "stress test with gemini", "what would gemini say", "cross-model review",
  or "get another opinion". Voice aliases: "gem", "gemini check". Especially
  useful for Google Ads changes, SEO metadata updates, campaign structure
  decisions, keyword strategies, and bid/budget changes — Gemini has native
  Google ecosystem knowledge that complements Claude's analysis.
triggers:
  - gemini
  - gemini review
  - gemini challenge
  - gemini consult
  - ask gemini
  - second opinion gemini
  - stress test gemini
  - gem review
  - gem consult
---

# Gemini — Cross-Model Second Opinion

You are orchestrating a cross-model review by launching Google's Gemini CLI as
an independent reviewer. Gemini brings native Google ecosystem knowledge —
especially valuable for Google Ads, Search Console, and SEO decisions where
Google's own AI has deeper context about how their platforms work.

**Unlike the code-only review pattern**, this skill handles three types of
changes:

1. **Code changes** — diffs, new files, refactors
2. **Google Ads changes** — campaign structure, bid strategies, keyword lists, negative keywords, ad copy, budget allocation
3. **SEO metadata changes** — title tags, meta descriptions, schema markup, robots directives, sitemap updates, content rewrites

---

## Step 0 — Detect Gemini CLI

```bash
command -v gemini >/dev/null 2>&1 && echo "GEMINI_FOUND" || echo "GEMINI_NOT_FOUND"
```

**If `GEMINI_NOT_FOUND`:** Stop and tell the user:

> Gemini CLI is not installed. Install it with:
>
> ```
> npm install -g @google/gemini-cli
> ```
>
> Then run `gemini` once to authenticate with your Google account, and retry.

**If `GEMINI_FOUND`:** continue silently.

---

## Step 1 — Detect Mode

Parse the user's request to determine the mode. Match against these patterns:

| Mode | Trigger phrases |
|------|----------------|
| **review** | "review", "check", "look at", "pass/fail", "gate", "approve" |
| **challenge** | "challenge", "stress test", "break", "adversarial", "find holes", "poke holes" |
| **consult** | "consult", "ask", "what does gemini think", "opinion", "advice", "strategy" |

**If ambiguous:** default to **review** for changes that exist in the diff, or
**consult** if the user is asking a question with no pending changes.

---

## Step 2 — Detect Change Type

Determine what kind of changes are being reviewed. Check in this order:

### 2a — Check for Google Ads changes

Look for signs of Ads-related work in the current conversation context:
- Recent MCP tool calls to universal `mcp__NotFair__google_ads_*` tools or a supported dedicated Google Ads namespace
- Discussion of campaigns, keywords, bids, budgets, ad copy, negative keywords
- Files like `.notfair/change-log.json` or Ads-related config changes

If found, set `CHANGE_TYPE=google-ads`.

### 2b — Check for SEO metadata changes

Look for:
- Recent calls to SEO skills (seo-analysis, meta-tags-optimizer, schema-markup-generator)
- Discussion of title tags, meta descriptions, schema markup, robots.txt, sitemaps
- Content rewrites or keyword targeting changes
- CMS content updates (Strapi, WordPress, etc.)

If found, set `CHANGE_TYPE=seo`.

### 2c — Check for code changes

```bash
git diff --stat HEAD 2>/dev/null || echo "NO_GIT_DIFF"
```

If there's a diff, set `CHANGE_TYPE=code`.

### 2d — Mixed or unclear

If multiple types are present, set `CHANGE_TYPE=mixed`. If nothing is found and
mode is **consult**, set `CHANGE_TYPE=consult-only`.

---

## Step 3 — Build the Context

Assemble the context payload that Gemini will review. Tailor it to the change type.

### For `google-ads` changes:

Summarize the proposed Ads changes in a structured block:

```
GOOGLE ADS CHANGE SUMMARY
==========================
Account: [account name/ID if known]
Change type: [campaign creation | bid adjustment | keyword changes | negative keywords | ad copy | budget | targeting | etc.]

BEFORE (current state):
[Describe current campaign/keyword/bid state]

AFTER (proposed changes):
[Describe what will change]

BUSINESS CONTEXT:
[Goal of the change — CPA target, ROAS goal, traffic objective, etc.]
```

### For `seo` changes:

```
SEO CHANGE SUMMARY
==================
Site: [URL]
Change type: [title tags | meta descriptions | schema markup | content rewrite | robots.txt | sitemap | etc.]

BEFORE (current state):
[Current metadata/content]

AFTER (proposed changes):
[New metadata/content]

TARGET KEYWORDS:
[Keywords being targeted, if applicable]

SEARCH INTENT:
[Informational / navigational / commercial / transactional]
```

### For `code` changes:

```bash
BRANCH=$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo "unknown")
DIFF=$(git diff HEAD 2>/dev/null)
STAT=$(git diff --stat HEAD 2>/dev/null)
```

Combine the diff stat and full diff into the context.

### For `mixed` changes:

Combine all applicable sections above.

---

## Step 4 — Run Gemini

Build and execute the Gemini CLI command based on mode and change type.

### Review Mode

```bash
gemini -p "You are a senior reviewer with deep expertise in Google's advertising platform, Google Search, and SEO best practices. You are reviewing proposed changes for correctness, effectiveness, and potential risks.

CHANGE TYPE: ${CHANGE_TYPE}

${CONTEXT}

Evaluate these changes and produce a structured review:

1. VERDICT: PASS or FAIL (use FAIL if any blocking issue exists)

2. BLOCKING ISS

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

gemini -p "You are a senior reviewer with deep expertise in Google's advertising platform, Google Search, and SEO best practices. You are reviewing proposed changes for correctness, effectiveness, and

Gates applied: no_behavioural_pass.