unisound-initial-record
ReviewAudited by ClawScan on May 15, 2026.
Overview
This skill coherently generates structured initial medical records from user-provided dialogue, but users should notice that sensitive medical text is sent to an external LLM API and optional debug/output files can persist data.
Install only if you are allowed to send the selected medical dialogue to the configured Hivoice medical LLM endpoint. Deidentify patient text first, protect the required app key, review the shared `_shared/doc-preprocess` dependency, and avoid `--save-prepared` unless storing preprocessed medical data locally is acceptable.
Findings (4)
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.
Medical dialogue may include sensitive health information, so any non-deidentified input will be transmitted to the configured LLM provider.
The skill clearly discloses that it sends the medical dialogue-derived prompt to an external/provider LLM endpoint using bearer authentication.
外部 API - 内部医疗大模型：`https://maas-api.hivoice.cn/v1/chat/completions` - 方法：POST，OpenAI 兼容格式 - 需要传入 `--appkey` 参数进行 Bearer 认证
Use only deidentified or approved-for-sharing medical text, confirm the provider endpoint is trusted, and understand the provider's data handling policy before use.
Anyone with the app key may be able to consume the associated LLM service quota or access permitted model functionality.
The script requires a bearer app key to access the medical LLM service. This is purpose-aligned, but it is still a credential users should scope and protect.
parser.add_argument(
"--appkey",
required=True,
help="必须传入。内部医疗大模型鉴权 key，使用 Bearer 方式认证。",
)Use a dedicated, least-privilege app key for this service, avoid sharing it in logs or command history, and rotate it if exposure is suspected.
The unreviewed shared preprocessing module will handle input files before they are sent to the LLM, so its trustworthiness matters.
The runtime imports preprocessing code from a shared skill directory that is not included in the supplied file manifest, so that dependency's behavior is outside this review.
PREPROCESS_DIR = SKILLS_ROOT / "_shared" / "doc-preprocess" / "scripts"
if str(PREPROCESS_DIR) not in sys.path:
sys.path.insert(0, str(PREPROCESS_DIR))
from preprocess import (Install `_shared/doc-preprocess` only from a trusted source and review or pin that dependency when using this skill with sensitive documents.
If `--save-prepared` is used, preprocessed medical dialogue can be stored on disk and may need the same protection as the original record.
The privacy section says input and intermediate results are not persisted, but the code provides an explicit debug option that saves the prepared payload. This is user-controlled and documented, but users should notice the exception.
if args.save_prepared:
save_prepared(payload, args.output, input_path)Do not use `--save-prepared` on identifiable patient data unless local storage is approved; delete debug files when no longer needed.