Security audit

Simple CSC

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent Chinese spelling-correction guide, but it sends users to run unpinned third-party setup scripts and dependencies that were not included in the reviewed artifact.

Review the upstream simple-csc repository and its scripts before running them. Prefer a pinned commit, pinned package versions, and an isolated environment without production credentials or broad host filesystem access. The localhost API examples are expected for local correction use, but avoid sending sensitive text unless you control the running model and server environment.

A.I.G

Vulnerability Patterns

Insecure DependenciesIntroduces malicious components through unsafe dependency sources
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)

T08 · Insecure Dependencies

Warning

Location: SKILL.md:24
Finding: Execution of Unpinned Third-Party Repository Scripts and Dependencies
Content: View full analysis
Remediation: View remediation
``` 2. Verify the selected commit using a trusted signed release or an independently distributed expected commit hash. 3. Review and vendor the required setup and download scripts inside the audited artifact, or publish their expected cryptographic checksums. 4. Use a dependency lockfile with exact versions and hashes. For direct installation, require a reviewed version and hash rather than: ```bash pip install flash-attn --no-build-isolation ``` 5. Pin HuggingFace models to immutable revisions and verify model files where checksums are available. 6. Require checksums or signatures for downloaded datasets and archives before parsing or extracting them. 7. Execute third-party setup and build operations in an isolated container or disposable virtual machine with: - No production credentials - No unnecessary host mounts - A non-root user - Restricted outbound network access - Minimal filesystem permissions 8. Inspect shell scripts and package build metadata before execution, particularly for network downloads, credential access, shell expansion, and lifecycle hooks. ]]>

SkillSpector

Vulnerability Patterns

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

Findings (3)

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The manifest says to trigger when a user asks to correct Chinese text spelling errors, set up the environment, work with datasets, or when code merely imports lmcsc or references LMCorrector. This spans many common repository tasks without clear boundaries or exclusion conditions, making activation ambiguous and likely to overlap with routine discussion of Chinese NLP work.
Content: No source excerpt is available for this finding.

External Transmission

Medium

Category: Data Exfiltration
Confidence: 60% confidence
Finding: Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
Content: Scanner excerpt · SKILL.md (reported line 108)May include surrounding context.
bash
# Non-streaming curl -X POST 'http://127.0.0.1:8000/correction' \ -H 'Content-Type: application/json' \ -d '{"input": "完善农产品上行发展机智。"}'

Natural-Language Policy Violations

Low

Category: Not specified by scanner
Confidence: 76% confidence
Finding: The API request schema uses the example value "要纠正的文本" for the input field, which implicitly assumes Chinese-language usage. Because this is natural-language documentation and no explicit user language choice or locale limitation is stated nearby, it may conflict with a policy requiring language/locale neutrality unless the skill is clearly documented as Chinese-specific.
Content: No source excerpt is available for this finding.