Back to skill

Security audit

Openclaw I18n Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is not overtly malicious, but it needs review because it can silently rewrite every response and its install path includes unnecessary unpinned Python dependencies.

Install only if you are comfortable with a skill that can persist language preferences and rewrite all agent output before you see it. Prefer using it in a constrained or opt-in way, avoid relying on it for code, commands, legal/medical/financial text, exact quotations, or multilingual content, and avoid the manual pip install path unless dependencies are pinned or removed.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T08 · Insecure Dependencies

Warning
Location
processor/requirements.txt:1
Finding

Unpinned and Unused Third-Party Dependencies Create Supply-Chain Exposure

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
processor/i18n_processor.py:31
Finding

Mandatory Context-Free Output Rewriting Can Corrupt Commands, Code, and User-Visible Content

Content
View full analysis
0: prev_code = ord(text[i - 1]) before_latin = (0x0041 <= prev_code <= 0x007A) or (0x00C0 <= prev_code <= 0x024F) before_non_latin = not before_latin and not (text[i - 1] in ' \t\n\r') is_stray = _is_stray_char(char, text, i) if is_stray a ...[truncated 4967 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (55)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The code generally fits a post-processing i18n cleaner, including diacritic correction and locale-specific cleanup. However, key declared capabilities are absent: there is no language auto-detection, and no date or currency formatting logic. Additionally, the declared description says Romanian and German are supported at launch, but the code also supports French, Spanish, and English passthrough. These are material differences between the declared description and the implemented behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The code snippet is minimal and only exposes/imports language submodules. It does not demonstrate the main declared behaviors such as language auto-detection, diacritic enforcement, date/currency formatting, or output cleaning. Additionally, the description says Romanian and German are supported at launch, while this chunk includes French and Spanish modules as well. Because the actual code shown is a package initializer rather than the described functionality, the supplied description is not accurately represented by this code chunk alone.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The declared description promises a broader i18n/l10n system with language auto-detection, locale formatting for dates and currency, and support focused on Romanian and German. The supplied code chunk instead implements low-level text cleaning helpers: stripping likely stray non-Latin characters, whitespace normalization, limited Romanian diacritic fixes, and dictionary-based typo correction. While this could be part of a post-processing cleaner mentioned in the description, it does not substantiate the larger claimed capabilities like auto-detection or locale formatting. Additionally, the code explicitly contains handling logic for many non-Latin scripts, which is not reflected in the declared scope. Therefore the description materially overstates and mischaracterizes what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code chunk is narrowly focused on German diacritic and spelling normalization using hardcoded dictionaries and regex replacements. While this partially aligns with the description's mention of enforcing correct diacritics and German support, it does not substantiate the broader declared purpose: there is no language detection, no Romanian implementation, no locale-aware date or currency formatting, and no evidence of a general i18n/l10n layer or output-cleaning pipeline. Because the actual code's primary purpose is much narrower than the declared description, this is a material description-behavior mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The description presents a broad localization subsystem with language detection, formatting features, and Romanian/German support. The supplied code is much narrower: a single Spanish module containing a typo dictionary and regex replacements for Spanish accents/diacritics. That is related to text cleanup/localization, but it materially differs from the declared primary scope and supported languages. There is no sign here of auto-detection, date/currency handling, or Romanian/German implementation.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied code chunk is narrowly focused on French typo and diacritic correction via hardcoded replacements and regex substitutions. That is only a small subset of the declared functionality. The declared description emphasizes a broader i18n/l10n layer with language auto-detection, locale-specific date/currency formatting, post-processing cleanup, and launch support for Romanian and German. In contrast, this file adds French-specific behavior, which is materially inconsistent with the stated supported languages, and none of the broader localization features are present in this chunk. This is a meaningful description-behavior mismatch rather than a harmless implementation detail.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared description presents a broader internationalization/localization layer with auto-detection, locale-aware formatting, and Romanian/German support. The supplied code only implements Romanian-specific post-processing: regex-based diacritic/typo correction and removal of stray CJK/Cyrillic/Arabic characters in Romanian text. The cleaning behavior is consistent with the 'post-processing cleaner' portion of the description, but key declared capabilities—language detection, date/currency formatting, and German support—are absent from this chunk. Therefore the description does not accurately represent what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied code is only a test file, so conclusions are limited to behaviors exercised by the tests. Those tests clearly support a post-processing cleaner that fixes Romanian/German diacritics, common typos, stray characters, and whitespace. However, the declared description also claims automatic language detection and locale-aware date/currency formatting, and none of those capabilities are evidenced here. The primary observed behavior is narrower text normalization rather than a full internationalization/localization layer. Therefore the description overstates what this code chunk actually demonstrates.

Content

No source excerpt is available for this finding.

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
90% confidence
Finding

The phrase 'Always respond' is an anti-refusal pattern because it attempts to impose unconditional behavior on the agent, which can conflict with higher-priority safety constraints or context-specific refusals. Even though the subject is language choice rather than overtly harmful content, rigid unconditional instructions are dangerous in prompt-based systems because they erode the agent's ability to decline or qualify unsafe requests.

Content

Scanner excerpt · SKILL.md (reported line 82)May include surrounding context.

md
## Output Rules

### Language Enforcement
- **Always respond in the detected or configured language**
- Never switch languages mid-response without explicit reason
- If a user writes in Romanian, respond in Romanian — never English
- If a user writes in German, respond in German — never English

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · changelog.html (reported line 38)May include surrounding context.

html
<h4 style="margin-top: var(--space-4); margin-bottom: var(--space-2);">Added</h4>
        <ul class="changelog-features">
          <li>SKILL.md — Agent configuration layer with language detection and output rules</li>
          <li>Post-processor engine — Python-based text cleaning with diacritics fixing</li>
          <li>Romanian support — ă, â, î, ș, ț diacritics, 30+ typo corrections</li>
          <li>German support — ü, ö, ä, ß diacritics, 20+ typo corrections</li>

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · changelog.html (reported line 10)May include surrounding context.

html
<link rel="stylesheet" href="style.css">
</head>
<body>
  <!-- Navigation -->
  <nav class="nav">
    <div class="nav-container">
      <a href="index.html" class="nav-logo">

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · docs.html (reported line 10)May include surrounding context.

html
<link rel="stylesheet" href="style.css">
</head>
<body>
  <!-- Navigation -->
  <nav class="nav">
    <div class="nav-container">
      <a href="index.html" class="nav-logo">

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · index.html (reported line 11)May include surrounding context.

html
<link rel="stylesheet" href="style.css">
</head>
<body>
  <!-- Navigation -->
  <nav class="nav" id="nav">
    <div class="nav-container">
      <a href="index.html" class="nav-logo">

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · index.html (reported line 81)May include surrounding context.

html
</div>
  </section>

  <!-- Features -->
  <section class="features">
    <div class="container">
      <h2 data-animate="fade-up" class="split-text">

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The function documentation explicitly says ambiguous Spanish words like 'si' must never be rewritten without context, yet the regex rule unconditionally replaces every standalone 'si' with 'sí'. This can silently alter the semantic meaning of user content, prompts, commands, or policy text, creating integrity issues and potentially changing downstream agent behavior.

Content

No source excerpt is available for this finding.

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · processor/languages/ro.py (reported line 117)May include surrounding context.

python
NEVER change:
    - Words where "s" or "i" is part of compound not separated
    - Unclear contexts — leave as-is
    """
    result = text

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · SKILL.md (reported line 79)May include surrounding context.

md
---

## Output Rules

### Language Enforcement
- **Always respond in the detected or configured language**

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · skill/SKILL.md (reported line 79)May include surrounding context.

md
---

## Output Rules

### Language Enforcement
- **Always respond in the detected or configured language**

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
78% confidence
Finding

The phrase 'Always respond' is an absolute behavioral directive that reduces the agent's flexibility to refuse, de-escalate, or switch modes when higher-priority safety or policy constraints apply. While not overtly malicious here, unconditional instructions are a common prompt-level anti-refusal pattern because they encourage compliance even when exceptions are necessary.

Content

Scanner excerpt · skill/SKILL.md (reported line 82)May include surrounding context.

md
## Output Rules

### Language Enforcement
- **Always respond in the detected or configured language**
- Never switch languages mid-response without explicit reason
- If a user writes in Romanian, respond in Romanian — never English
- If a user writes in German, respond in German — never English

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This stylesheet is a full marketing/landing-page theme with hero sections, pricing cards, testimonials, docs/changelog layouts, and waitlist UI, which is materially inconsistent with an i18n/localization skill. In a skill package, large off-purpose components increase supply-chain risk by hiding unexpected functionality, expanding review surface, and creating opportunities for deceptive bundling or future abuse even if the CSS itself is not executable code.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The README states that the skill configures the agent to think in the 'correct language' and emphasizes auto-detection and enforcement, which can override explicit user preference or mis-handle ambiguous input. In an agent setting, forced language behavior can reduce user control, cause incorrect communication, and interact dangerously with the downstream cleaner that rewrites responses automatically.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The README explicitly describes a post-processing layer that mechanically modifies every model response by fixing diacritics, removing characters, and normalizing text before the user sees it. Even if intended for quality improvement, silent transformation of all outputs can change meaning, remove legitimate content, or mask model behavior, especially in multilingual or technical contexts where characters and spacing matter.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Mandating responses in the detected/configured language without clear user confirmation can override user intent and create unsafe misunderstandings, especially in multilingual or safety-critical interactions. In agent settings, forced language selection can also be abused to suppress clarification, reduce transparency, or make policy/safety content less accessible to the user.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Transparent modification of every model response before delivery creates an integrity risk because the user cannot distinguish original model output from post-processed text. A silent cleaner can inadvertently alter meaning, redact content, or introduce factual and safety errors while leaving no visible audit trail.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
80% confidence
Finding

The documentation describes a system that modifies model output to fit a detected or active language, which can override or transform user-intended content without clearly preserving user choice. In an agent context, silent post-processing of outputs can degrade integrity, misrepresent user meaning, or cause unsafe alterations to quoted, technical, or multilingual text.

Content

No source excerpt is available for this finding.