llm-text-correct
Security checks across static analysis, malware telemetry, and agentic risk
Overview
The skill mostly matches Chinese text correction, but it automatically changes Python environments, installs broad unpinned dependencies, and may persist corrected private text in logs.
Install only if you are comfortable with it creating or changing Python environments and downloading large ML dependencies/models. Prefer running it in an isolated environment, remove or pin unnecessary dependencies, avoid broad/private folders and sensitive pasted text, and review generated files and logs.
Static analysis
No static analysis findings were reported for this release.
VirusTotal
VirusTotal findings are pending for this skill version.
Risk analysis
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Invoking the skill may download and run a large dependency set, increasing supply-chain and local compatibility risk.
The runtime setup installs broad, unpinned dependency chains, including audio packages that are not used by the Chinese text-correction workflow.
subprocess.check_call([str(venv_python), "-m", "pip", "install", "torch", "torchvision", "torchaudio", "--index-url", index_url]) ... subprocess.check_call([str(venv_python), "-m", "pip", "install", "audio-separator[gpu]", "librosa"])
Use a reviewed install spec or lockfile with pinned versions, remove unused audio dependencies, and clearly disclose expected package and model downloads.
The skill can modify the active Python environment automatically, not just process text.
A top-level module call can run pip and force-reinstall packaging tools as soon as the helper is imported, before the script's environment handling is clearly contained.
fix_setuptools_for_legacy_packages() ... subprocess.check_call([sys.executable, "-m", "pip", "install", "--quiet", "--force-reinstall", "setuptools<=81.2.0", "wheel"])
Move environment changes into an explicit, user-approved setup step and avoid modifying the host interpreter on import.
A failed or incompatible install could break other skills or workflows that depend on the same shared Python environment.
The virtual environment is intentionally placed in a shared parent skills directory, so this skill's dependency changes can affect other components that reuse the same environment.
VENV_DIR = _BASE_PATH.parent.parent.parent / "venv" #venv虚拟环境目录在skills/venv通用目录
Use a per-skill isolated virtual environment or document and obtain approval for shared-environment changes.
If pointed at a large or private folder, the skill will read all matching files and create a parallel corrected output tree.
Folder mode recursively reads many text and code file types. This is user-directed and writes corrected copies, but it is broader than simple pasted-text correction.
extensions = {".txt", ".md", ".py", ".json", ".html", ".csv", ".js", ".css"}
return [p for p in folder.rglob("*") if p.is_file() and p.suffix.lower() in extensions]Provide narrow file or folder paths and review the generated corrected files before using them.
Private documents, chats, or other sensitive text may remain on disk after correction.
In direct-text mode the full corrected output is sent to the skill logger, which can retain sensitive pasted text in local logs.
corrected = correct_text(args.input, use_refine=use_refine) logger.info(corrected)
Avoid using sensitive text until logging is minimized or redacted, and document log location and retention for users.