blog-polish-zhcn-images

• Outcome: Review
• Latest audit: Aug 8, 2026 · 6:21 AM UTC
• Version: 1.0.7

Overview

The skill is framed as a real blog polishing, translation, and image-prompt generator, but its executable workflow mostly saves the original draft with a simulated translation note and hard-coded image prompts.

Review before installing if you need reliable publication-ready translation: this skill may leave the source text essentially unchanged while reporting a generated package. It is best treated as a draft/demo workflow unless the implementation is fixed to actually translate, polish, and derive prompts from the article. Avoid using sensitive drafts unless you are comfortable with persistent files being saved under ~/.openclaw/workspace/contentPolished/.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Rogue Agent: Self-Modification, Session Persistence
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Description-Behavior Mismatch

High

• Confidence: 98% confidence
• Finding: The skill advertises real polishing and zh-CN translation, but the implementation only prepends a header, appends a note, and writes the original draft unchanged. This is a security-relevant integrity issue because downstream users or agents may trust the output as transformed content and publish incorrect, untranslated, or policy-noncompliant material.

Description-Behavior Mismatch

High

• Confidence: 97% confidence
• Finding: The image prompt step generates hard-coded filenames and prompts without inspecting the draft or honoring user-provided style, background, or aspect ratio inputs. This breaks the contract of the skill and can mislead callers into believing prompts were derived from article sections and user preferences when they were not.

Intent-Code Divergence

Medium

• Confidence: 90% confidence
• Finding: The implementation explicitly says it is simulating translation while the broader skill documentation presents the capability as real. This discrepancy can cause unsafe reliance on outputs in automated workflows, especially where language conversion or content review is expected before publication.

Session Persistence

Medium

• Category: Rogue Agent

Content

- code blocks and commands (no translation inside code fences)
  - inline code spans (backticks) as-is unless obviously wrong
- **Images**: generate a **hero prompt** + **one prompt per section**, all in the same style/tone
- **Save**: write everything to `~/.openclaw/workspace/contentPolished/` using `yymmddhhmm-*` naming

## When to Use This Skill

• Confidence: 88% confidence
• Finding: The skill reads from and writes to persistent locations under the user's home directory, causing content to remain across sessions and potentially be reused or exposed unintentionally. In agent environments, persistent storage of drafts and polished outputs can leak sensitive or unpublished material to later tasks, users, or tools with filesystem access.

Session Persistence

Medium

• Category: Rogue Agent

Content

1. Decide `subject`:
   - `subject = input.subject || slugify(title)` (lowercase, hyphens)
2. Write the polished markdown:
   - `{{outputDir}}/{{ts}}-{{subject}}.md`
3. Determine image filenames:
   - Hero: `{{ts}}-main.png`

• Confidence: 90% confidence
• Finding: The save step persists polished content and image filenames to a durable output directory using predictable naming. This increases the risk of unintended disclosure of sensitive drafts, especially if multiple sessions share the same workspace or if other tools automatically ingest files from that location.

VirusTotal

64/64 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.