Back to skill

Security audit

social-overlays

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because it reports presets that can create synthetic viewer counts and unpinned commands that may execute changing upstream releases; scan coverage is incomplete.

Consider avoiding installation until the viewer-count behavior is limited to clearly labeled demos or uses real data, and the HyperFrames commands are pinned to a reviewed version. Coverage was marked partial: a referenced local path was missing, and bundled font files were outside some analyzer checks.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

LIVE badge preset specifies synthetic viewer-count growth despite the real-data-only manifest.

High
Category
Excessive Agency
Confidence
98% confidence
Finding

The preset specifies percentage-based, seeded viewer-count growth rather than requiring observed viewer measurements. Its allowance for illustrative counts does not require an on-screen disclosure or isolate synthetic behavior to demos, so applying the preset to ads can produce fabricated social proof despite the real-data-only policy.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx hyperframes'.

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The documented npx command does not pin an exact package version, so it can download and execute a changing release when no local installation exists. A compromised upstream release could execute code with the agent's permissions; the prose reference to HyperFrames 0.8 does not constrain package resolution. This is a dependency supply-chain risk, not evidence of an MCP server.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx hyperframes'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The catalog command invokes an unversioned npm executable, potentially fetching and executing the current upstream release. Although catalog discovery is a legitimate workflow, a compromised release would run with the agent's access to local files and credentials. The MCP classification is inaccurate, but the unpinned execution risk is real.

Content

No source excerpt is available for this finding.

Review-platform preset requires an official logo despite the file's explicit no-logos documentation.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

In assets/presets/review-proof-card.json, L100–L108 specify an official logo and unmodified official assets for the review-platform preset. This actively contradicts the same file's trademark_note at L185, which describes the presets as generic inspired looks with no logos, making the documented branding boundary unreliable.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx hyperframes'.

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

The preview command omits the exact version used by the preceding check and render commands. Those pinned invocations do not guarantee a persistent local installation, so preview can resolve a different upstream release and execute compromised code with the agent's permissions. This concerns npm execution rather than an MCP server.

Content

No source excerpt is available for this finding.

LIVE viewer-count implementation synthesizes growth instead of requiring real viewer data.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The implementation directs the builder to synthesize increasing viewer counts from a seed and starting value without requiring a real viewer series. Warnings against fake counts do not resolve this operational contradiction, and an 'example' note only in the brief does not inform viewers. In commercial overlays, the resulting numbers can misrepresent audience engagement.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/photo-wall-loop.md (reported line 42)May include surrounding context.

md
- Jump at the loop point → speeds must be k × H / D [29].
- Same speed for every column → vary k for depth.
- Using random cards order that repeats a photo side-by-side → shuffle with a fixed seed and check neighbours.
- Customer posts used without consent or faked "UGC" → real, permissioned content [21].
- Headline floating over busy cards with no scrim → 50% scrim or blur band [42].
- 25 px/frame scroll strobing → keep ≤ ~12 px/frame.