T09 · Insecure Skill Coding Practices
Warning
- Location
SKILL.md:116- Finding
Automatic Unconsented Persistence of Document-Derived Metadata
- Content
View full analysis
Vulnerability Details
File Location:
SKILL.md, lines 116-132 and 160-171
Vulnerability Type: Unrequested persistent local database write
Risk Level: MediumEvidence
The following is an English rendering of the complete relevant Skill directives:
markdown ### Supplement to Section Four: Extract table layouts and automatically store them in a database When analyzing a paper, proactively identify every table and extract the following formatting logic for each table: 1. Table name: Table 1, Table 2, and so on 2. Row structure: how variables are arranged in rows, including nesting and grouping 3. Column structure: relationships among columns, such as different models or exposure groups 4. Variable order: ordering rules within rows, such as demographics, lifestyle, and metabolic indicators 5. Formatting conventions: - Continuous variables: mean ± SD or median (IQR) - Categorical variables: n (%) format and decimal precision - HR/CI: decimal precision and reference-group notation 6. Statistical methods: tests and models used 7. Footnote and annotation conventions Storage path: `D:\autoclaw\results\medical-research-methods-library\methods.db` Database table: `table_layouts`markdown ## Section Eight: Table Layout Conventions (Automatically Stored in the Database) - Table-layout extraction is performed automatically. - User authorization is not required. - During paper analysis, the extracted information must be written to the database.Technical Analysis
The Skill mandates a persistent filesystem side effect that is not necessary to provide the requested paper analysis. It automatically stores document-derived metadata in a hardcoded absolute SQLite database path without requiring informed user consent.
The instructions do not define:
- Whether the submitted paper is confidential or contains sensitive research information.
- Which precise values ma ...[truncated 2562 chars]
- Remediation
View remediation
Remediation Suggestions
- Remove mandatory automatic database writes from the default analysis workflow.
- Require explicit, informed, per-request user consent before retaining any document-derived information.
- Present the exact destination, fields, retention period, and deletion procedure before requesting consent.
- Default to returning table-layout information in the response without persistent storage.
- Use a user-selected or application-managed sandbox directory instead of a hardcoded absolute path.
- Validate that the destination database is owned by the expected application and has the expected schema before writing.
- Use parameterized SQL statements, explicit transactions, rollback handling, and restrictive file permissions.
- Minimize stored data and avoid retaining paper titles, authors, unpublished results, or other identifying information unless specifically authorized.
- Add record provenance, retention limits, user inspection controls, and a reliable deletion mechanism.
- Clearly report whether persistence succeeded and never silently fall back to another path.