ClawHub
Back to skill

Security audit

Skills Search across ClawHub, skills.sh, and SkillsMP

Security checks for vulnerabilities and agentic risk

Overview

This skill is coherent for finding skills, but needs review because it sends search terms to outside services and includes shell/install command templates that can be unsafe if followed blindly.

Install only if you are comfortable with skill-search queries being sent to ClawHub, skills.sh, SkillsMP, and the r.jina.ai proxy. Avoid using sensitive internal names as search terms. Review any recommended install command before running it, especially npx commands and commands ending in --yes, and prefer pinned or trusted package sources. Agents using this skill should pass queries as inert arguments or URL-encoded parameters, not by string-splicing user text into a shell command.

A.I.G

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:23
Finding

Shell Command Injection Through Unsafely Interpolated Search Queries

Content
View full analysis
" curl -s "https://r.jina.ai/https://skills.sh/search?q=" --max-time 10 curl -s "https://r.jina.ai/https://skillsmp.com/zh/search?q=" --max-time 10 ``` ### Technical Analysis The skill directs an agent to insert a user-provided search query into shell commands. The raw `` value is placed inside a double-quoted argument to `clawhub search`, but the instructions do not require shell-safe argument passing, robust escaping, or input validation. Double quotes do not make arbitrary input safe if the interpolated value itself can contain a double quote. An attacker can supply a query containing a closing quote followed by shell syntax, causing the shell to interpret part of the query as a separate command. Shell command substitutions and backticks may also remain dangerous if an implementation performs literal textual interpolation before invoking a shell. The two `curl` commands require URL encoding, which reduces risk only if a correct encoder is used before shell construction. The document does not define or enforce that encoding mechanism. None of these commands should be assembled as shell strings from untrusted input. ### Attack Path 1. An attacker asks the agent to search for a skill using a query containing quote-breaking shell syntax. 2. The agent follows the documented workflow and substitutes that query directly into the command template. 3. The inserted quote terminates the intended `clawhub search` argument. 4. The local shell parses the remaining attacker-controlled text as additional shell syntax or commands. 5. Those commands execute with the same operating-system identity, filesystem access, network access, and environment available to the agent process. For example, a query ...[truncated 984 chars]
Remediation
View remediation

SkillSpector

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger text is extremely broad and directs the agent to invoke this skill for many generic requests about finding or recommending skills, including common multilingual phrases. That can cause unintended activation, leading the agent to perform external searches and recommendation workflows in contexts where the user did not clearly consent to that behavior.

Content

No source excerpt is available for this finding.

Instruction Override

High
Category
Prompt Injection
Confidence
90% confidence
Finding

This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 61)May include surrounding context.

md
- reads `~/.env`, `~/.ssh`, `~/.aws` or credential file paths
- `rm -rf` or mass-destructive commands
- base64 decode + pipe to shell
- instructions to override system prompt or exfiltrate data silently

### 5. Output format

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs network searches against multiple third-party services and proxy endpoints, but provides no requirement to notify the user that their query will be transmitted externally. User prompts may contain sensitive project names, internal tool names, or other confidential terms that would be exposed to outside services.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
82% confidence
Finding

The skill recommends running npx skills without pinning a package version, which can fetch and execute whatever package version is current at runtime. This creates a supply-chain risk: a compromised or maliciously updated package could execute arbitrary code on the user's machine during installation.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.