ℹ
Purpose & Capability
The name/description (extract diagrams/prompts from local medical papers and save to a local file) aligns with the instructions and the declared mcp capability in config.yaml. However, the SKILL.md hard-codes the local paths to /home/xie/..., which is specific to a particular user environment and not generally appropriate. Requiring Gemini (external service) and MCP browser/file-write are plausible for the stated purpose but the hard-coded path reduces portability and suggests the skill was authored for a single host.
!
Instruction Scope
Instructions explicitly demand uploading a local medical paper to Gemini via MCP browser-file-upload and prohibit outputting parsed results until the skill writes them to disk. Uploading possibly sensitive medical PDFs to an external model/service is a privacy risk. The instructions also force procedural constraints (never end unless step 7 writes file, forbid printing intermediate results) that can cause the agent to insist on writing to disk or loop until the write succeeds. The requirement to use browser-file-upload rather than other UI actions is unusually prescriptive but consistent with automated file transfer—still it increases the chance of exfiltration to an external service.
✓
Install Mechanism
Instruction-only skill with no install steps and no additional binaries or packages declared. From an install-mechanism viewpoint this is low-risk—nothing is written to disk by an installer and there are no download URLs.
!
Credentials
The skill requests no environment variables or credentials, which is appropriate. However, it requires access to specific local filesystem paths (/home/xie/桌面/...), which is a form of privileged local access: the skill will read a local PDF and write an output file. For medical data this is a meaningful sensitivity concern. The fixed home path is disproportionate for a general skill and may indicate it was designed for a single user's environment rather than general use.
ℹ
Persistence & Privilege
The skill does not request always:true and does not modify other skills/configs. It does require using MCP to write a local file (normal for a tool that saves results). The coercive requirement 'task not complete until file written' and the ban on outputting results before write are unusual and can cause the agent to continue operating until the file write succeeds; this increases operational/risk surface (loops, repeated uploads) but is not itself an elevation of privilege.
What to consider before installing
Before installing or running this skill, consider the following:
- Privacy: the skill instructs uploading a local PDF (medical imaging paper) to Gemini (an external model/service). If the paper contains sensitive patient data or confidential research, do not upload it without appropriate consent and data‑handling controls.
- Hard-coded path: the skill uses /home/xie/桌面/... for both input and output. If your files are elsewhere the skill will fail or may attempt to access unrelated files. Ask the author to parameterize the base directories or confirm the intended host.
- Forced behavior: the SKILL.md forbids ending the task unless the agent successfully writes the output file and forbids printing results before writing. This can cause repeated uploads, long-running loops, or unexpected file writes—test carefully with a non-sensitive dummy PDF first.
- Access required: the skill needs MCP browser/file-upload and MCP file-write capabilities enabled. Review what MCP will actually do (network upload, write to your filesystem) and ensure you trust the endpoint.
- If you need to proceed: run a controlled test (dummy paper) and verify exactly what data is sent to Gemini and what ends up in /home/youruser/desktop/analysis/. If the skill is intended for general distribution, ask the maintainer to remove hard-coded paths, document privacy implications, and add explicit consent and configurable paths.
If you can provide more context—who operates /home/xie on the target machine, whether Gemini Pro is an approved service for your data, or whether the tool should run entirely locally—I can raise or lower the confidence and refine the advice.
