Back to skill

Security audit

SEO Intel

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent SEO tool, but it also guides agents toward cloud analysis and live Cloudflare deployment without a strong approval gate.

Install only if you are comfortable with website crawling, persistent local SEO data, and cloud-backed model analysis. Do not allow autonomous production deploys; require a diff, exact target confirmation, and explicit human approval before any Wrangler deploy. Pin package versions where possible and use an isolated environment without unrelated credentials for scans.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:20
Finding

Unpinned Global npm Package Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:20-23
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code

bash
npm install -g seo-intel
seo-intel setup      # detects OpenClaw automatically, configures everything

Related instances also recommend installing Wrangler without a pinned version:

  • SKILL.md:500-503
  • AGENT_GUIDE.md:282-285
bash
npm install -g wrangler
wrangler login        # opens browser for Cloudflare OAuth

Technical Analysis

The instructions install globally scoped npm packages using mutable package names without:

  • Pinning reviewed versions
  • Verifying package integrity hashes or signatures
  • Using a lockfile
  • Disabling or reviewing npm lifecycle scripts
  • Confirming the expected registry and package provenance

An npm package can execute lifecycle scripts during installation. Because no version is specified, the command resolves to whichever release the configured registry currently identifies as latest. Consequently, the code ultimately installed and executed can differ from what was reviewed during this audit.

The audited artifact contains documentation but does not include the source code of either external package. Their runtime behavior therefore cannot be validated from this project alone. This is a supply-chain exposure, not evidence that either named package is currently malicious.

Attack Path

  1. An attacker compromises the publisher account, package release process, or npm registry path for one of the named packages.
  2. The attacker publishes a malicious release or causes a malicious version to resolve as the latest package.
  3. A user or agent follows the documented npm install -g command.
  4. npm downloads the mutable release and may execute its installation lifecycle scripts.
  5. The malicious package runs under the installing user's privileges and remains available as a globally installed executable.
  6. If the compromised executab ...[truncated 891 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin every installed tool to a reviewed version, for example:

    bash
    npm install -g seo-intel@1.5.21
    npm install -g wrangler@<reviewed-version>
    
  2. Publish and verify expected package integrity hashes or signed release artifacts before installation.

  3. Explicitly configure the trusted registry and verify package ownership and provenance:

    bash
    npm config get registry
    npm view seo-intel@1.5.21 dist.integrity repository maintainers
    
  4. Prefer project-local, lockfile-controlled dependencies over global installation where practical.

  5. Review package contents and lifecycle scripts before installation. Where compatible with the package, initially install with lifecycle scripts disabled:

    bash
    npm install --ignore-scripts
    
  6. Run installation and crawling in an isolated, least-privileged environment without unrelated credentials.

  7. Keep Cloudflare authentication separate from analysis environments and require explicit human approval before live deployment.

  8. Add documented version-upgrade procedures requiring source review, integrity verification, and testing before changing the pinned versions.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (10)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The guide materially expands an SEO analysis skill into a change-and-deploy capability for third-party websites, including concrete instructions to edit site files and publish them globally via Wrangler. This violates least privilege: analysis tooling should not bundle operational deployment patterns, because an agent following this guide could make unauthorized production changes or turn untrusted analysis output into direct site modifications.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documentation introduces shell execution and production deployment (execSync, wrangler deploy, file copy commands) that are not justified by the stated purpose of competitive SEO intelligence. Embedding command execution patterns in an analysis skill increases the chance that an agent will perform high-impact local or production actions from loosely validated findings, creating a path to command misuse and unintended deployment.

Content

No source excerpt is available for this finding.

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
96% confidence
Finding

This command allows selecting an external model/provider for scan analysis, which implies potentially sending crawled content to third-party AI services. Without explicit consent, provider disclosure, and data-handling constraints, this creates a meaningful exfiltration and compliance risk for any scanned content.

Content

Scanner excerpt · SKILL.md (reported line 93)May include surrounding context.

seo-intel scan carbium.io # Full pipeline with AI-enriched export seo-intel scan carbium.io --no-ai # Deterministic export only (no LLM enrichment) seo-intel scan carbium.io --pages 50 # Limit crawl to 50 pages seo-intel scan carbium.io --model claude # Use Claude instead of Gemini seo-intel scan carbium.io --no-stealth # Disable stealth browser mode

text

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
95% confidence
Finding

The blog-draft command supports multiple external models, meaning project data and ledger-derived intelligence may be sent to outside providers during content generation. Because the command is presented as routine usage, users may not realize that proprietary strategy data or extracted site content is leaving the local environment.

Content

Scanner excerpt · SKILL.md (reported line 243)May include surrounding context.

seo-intel blog-draft # Auto-pick topic from ledger seo-intel blog-draft --topic "api security" # Specific topic seo-intel blog-draft --lang fi # Finnish seo-intel blog-draft --model claude --save # Use Claude, save to reports/

text

**Models:** gemini (default), claude, gpt, deepseek

Self-Modification

High
Category
Rogue Agent
Confidence
90% confidence
Finding

The skill includes instructions to copy and update its own published skill metadata and related public-facing files after releases, which is a form of self-modification of agent/tooling surfaces. In an automated or loosely supervised agent workflow, this can enable unreviewed propagation of changed instructions or drift across trust surfaces, increasing supply-chain and integrity risk.

Content

Scanner excerpt · SKILL.md (reported line 573)May include surrounding context.

After any version bump or feature release, update and redeploy. Also keep public listing surfaces aligned, not just local docs:

bash
# Update skill.md from source
cp /path/to/seo-intel/skill/SKILL.md /path/to/site/seo-intel/skill.md

# Update version references in llms.txt and llms-ctx.txt

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · AGENT_GUIDE.md (reported line 10)May include surrounding context.

javascript
import { run, capabilities, pipeline } from 'seo-intel/froggo';

// Run any command — always returns structured JSON
const result = await run('aeo', 'myproject');
// → { ok: true, command: 'aeo', project: 'myproject', timestamp: '...', data: { scores, summary } }

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The automatic analyze-fix-deploy flow presents deployment as a normal continuation of analysis and editing, but does not place an explicit confirmation or approval gate immediately before production deployment. In an agentic setting, this makes it more likely that inferred or incorrect edits will be pushed live without the user understanding the scope, target, or risk.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill advertises itself for a very wide range of common SEO, content, and website-analysis requests, which increases the chance of over-triggering in contexts where the user did not explicitly consent to crawling, competitor analysis, or downstream cloud processing. In an agent ecosystem, broad activation can cause unnecessary data access and tool use beyond user intent, especially because this skill can lead into external analysis and deployment workflows.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly states that extracted crawl data is analyzed by OpenClaw cloud models, but it does not present a clear, front-loaded warning that site content, competitor data, and derived analysis may be transmitted to external providers. This creates a real data-governance and confidentiality risk, especially if users run the tool on private staging sites, internal docs, or sensitive business content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documented Wrangler workflow enables immediate live deployment from analysis output, and while later text notes there is no staging or undo, the warning is not sufficiently prominent before the deploy sequence itself. In practice, an agent or user following this flow could push broken, unsafe, or unintended changes directly to production with little friction.

Content

No source excerpt is available for this finding.