Back to skill

Security audit

Side Hustle

Security checks for vulnerabilities and agentic risk

Overview

The skill is intended to search for side-hustle opportunities, but its real execution path is outside the package and its triggers and learned content are too broad and inconsistent.

Install only if you trust the local capability_executor and capabilities_clawhub.py environment that will execute this skill. Expect browser/web-search activity, and review or remove the broad/unrelated triggers and unspecified failure-reporting links before using it in a privacy-sensitive session.

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
Findings (1)

other

Warning
Location
SKILL.md:22
Finding

Execution Delegated to Unbundled and Unverifiable External Components

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 22-45
Vulnerability Type: Unverifiable external capability execution
Risk Level: Medium

Vulnerable Code

markdown
## Implementation
Calls `open_browser_and_search_side_hustle()` from `capabilities_clawhub.py`

## Usage
```python
from capability_executor import execute_capability
execute_capability("side hustle")

Dependencies

  • capabilities_clawhub.py in D:\coze-local\db
  • capability_executor.py (registration + dispatch)

Linkage

  • Reports failures to skill_evolver (MetaClaw) and seal_coevolution (SEAL)
  • Confidence tracked via capability_executor
text

### Technical Analysis

The Skill does not include its capability implementation or dispatcher. Instead, it delegates behavior to `open_browser_and_search_side_hustle()` in an external `capabilities_clawhub.py` file and invokes it through the external `capability_executor` dispatcher.

Because neither component is included in the audited package, their implementation, integrity, permissions, network destinations, and data-handling behavior cannot be verified. The Skill also declares that failures are reported to `skill_evolver` and `seal_coevolution`, but it does not document what information is transmitted, where it is stored, or whether user consent is required.

This creates an external trust boundary: the reviewed Skill text does not fully define the behavior that will occur when the trigger is invoked. An attacker or compromised local component able to modify the external Python modules could cause the legitimate-looking Skill invocation to execute behavior not represented in the reviewed package.

### Attack Path

1. A user activates the Skill using the configured side-hustle trigger.
2. The runtime calls `execute_capability("side hustle")` through the external `capability_executor` module.
3. The dispatcher resolves the trigger to `open_browser_and_search_side_hustle(
...[truncated 1362 chars]
Remediation
View remediation

Remediation Suggestions

  1. Include capabilities_clawhub.py, capability_executor.py, and all relevant registration code in the reviewed package so the complete execution path can be audited.
  2. Pin trusted versions or cryptographic hashes for external components and verify their integrity before capability registration or execution.
  3. Restrict the capability to the minimum required browser and network permissions, using an explicit allowlist of permitted destinations.
  4. Document every network request, telemetry destination, transmitted field, retention policy, and failure-reporting condition.
  5. Require explicit user consent before sending failure information or execution metadata to skill_evolver, seal_coevolution, or any other external subsystem.
  6. Remove evolution-system reporting if it is not essential to the advertised side-hustle search functionality.
  7. Validate capability registration so the trigger can resolve only to the expected, integrity-checked function.
  8. Execute the external capability in a sandbox with restricted filesystem, process, credential, and network access.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger scope includes very broad financial terms like '赚钱' and '兼职', which can cause the skill to activate during ordinary conversations about money or jobs rather than explicit requests to run this capability. Because the implementation opens a browser and performs searches, unintended invocation can lead to unexpected external actions, privacy leakage through search queries, and reduced user control.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill says it 'searches for side hustle opportunities' but does not clearly warn that invoking it will open a browser and perform live searches. This matters because browser launch and outbound search are side effects that can surprise users, expose sensitive context in queries, and violate expected consent if not disclosed up front.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest and top-level description say this skill searches for side hustle opportunities, but a large portion of the file documents Bilibili learning entries about unrelated songs, TV shows, videos, and media containing words like 'hustle' or 'side'. This is not merely incomplete documentation; it actively presents unrelated content as part of the skill's learned behavior and usage context, contradicting the intended function.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The catch-all condition '用户需要...相关帮助' is ambiguous and can match loosely inferred intent, expanding activation beyond explicit user requests. In a skill that launches browser-based searches, this weakens consent boundaries and increases the chance of unprompted or contextually incorrect external actions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

Lines describing trigger scenarios frame the skill as helping with earning money or part-time work, yet the subsequent learned examples are dominated by unrelated videos whose only connection is lexical overlap with 'hustle' or 'side'. This mismatch means the documentation implies a job/opportunity-search intent while showcasing behavior aligned with media keyword matching instead.

Content

No source excerpt is available for this finding.