Security audit
Security checks for vulnerabilities and agentic risk
This skill removes watermarks locally from selected documents, with disclosed output changes and no evidence of hidden access or data transmission.
Keep originals and inspect representative cleaned pages before relying on the results; automated pixel checks cannot guarantee perfect preservation. Header-logo removal is enabled by default and can be disabled with --no-logo. PDF output becomes raster images and loses selectable text. The verifier's older threshold documentation needs updating.
Without declared permissions the skill's intent is opaque and cannot be validated.
Both verification paths use a hard-coded 90% retention threshold rather than the documented 99% for images and 97% for PDFs, leaving the corresponding constants unused. Outputs below the promised retention thresholds can therefore receive 'ALL PASS' if the separate erased-content and residue checks also pass; those checks may limit the impact, but their implementations are not provided and cannot establish equivalent protection. This matters particularly for scanned notes and exam papers, where falsely approved content loss can remove meaningful information.
Overly Broad Trigger: description clause 'Use when the user asks to 去水印 / remove watermark / clean watermarks from 扫描笔记、讲义、试卷、资料图片、PDF 讲义, especially batches of course-note scans with' activates on 'to', which is too short and may match unintended inputs
No suspicious patterns detected.
