unisound-discharge-record
ReviewAudited by ClawScan on May 15, 2026.
Overview
The skill is largely purpose-aligned for generating discharge records, but it handles sensitive medical records and has a material privacy mismatch around local persistence.
Review this skill before installing. Use it only with approved, de-identified medical records and a trusted model endpoint. Protect the app key, verify the shared preprocessing dependency, and avoid `--save-prepared` or other local output options unless you intentionally want sensitive medical text written to disk.
Findings (5)
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.
Patient medical-record content leaves the local environment and is sent to the configured model service.
The script sends the generated prompt, which contains the patient record text, to a configured LLM endpoint using a Bearer credential.
DEFAULT_LLM_BASE = "https://maas-api.hivoice.cn/v1" ... headers = {"Authorization": f"Bearer {appkey}"} ... "messages": [{"role": "user", "content": prompt}]Use only an approved internal endpoint, de-identify records before use, and avoid overriding the base URL unless the destination is trusted.
A user may believe no medical text will be stored locally, while a debug workflow can create local files containing sensitive patient information.
The privacy statement says inputs and intermediate results are not persisted, but the same document describes a flag that saves preprocessed text locally.
- **不做本地持久化**：不将输入与中间结果写入本地持久化存储；**本次调用结束即销毁**。 ... - `--save-prepared`：可选。保存预处理后的文本，便于调试。
Clarify the privacy statement to state the exact exceptions, keep debug saving off by default, and warn users before saving prepared medical text.
Sensitive medical text may remain on the local filesystem after the run if debug saving is enabled.
When the documented save-prepared option is used, the preprocessed medical input is written to disk.
prepared_path = save_dir / f"{input_path.stem}.prepared.txt" ... prepared_path.write_text(payload_to_prepared_text(payload), encoding="utf-8")Avoid `--save-prepared` for real patient data unless storage is necessary, access-controlled, and covered by the user's data-retention policy.
Anyone with the app key may be able to call the internal model service under the user's authorization.
The skill requires a service credential for the medical-model API; this is purpose-aligned, but users should notice that a Bearer key is needed and should be protected.
`--appkey STRING`：**必填**。内部医疗大模型鉴权 key，使用 Bearer 方式认证。
Provide the app key only through a secure workflow, avoid putting it in shared logs or shell history, and rotate it if exposed.
Medical files may be processed by shared helper code that was not visible in this review artifact set.
The script imports a preprocessing module from a shared directory outside the included skill files; SKILL.md discloses this dependency, but the helper code is not present in the supplied manifest.
PREPROCESS_DIR = SKILLS_ROOT / "_shared" / "doc-preprocess" / "scripts" ... from preprocess import PreprocessError, SUPPORTED_FILE_TYPES, detect_input_type, load_input_artifact
Verify that `_shared/doc-preprocess` is installed from a trusted source and review or pin that dependency before using real patient data.