ma-publication-quality
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests manuscript files and analysis results to generate quality audit reports, which presents a surface for indirect prompt injection if the source content is maliciously crafted.
- Ingestion points: The skill reads manuscript files (e.g.,
07_manuscript/00_abstract.qmd,07_manuscript/03_results.qmd) and analysis outputs (e.g.,06_analysis/tables/*.csv) via scripts such asclaim_audit.pyandclaim_table_check.py. - Boundary markers: There are no explicit instructions or delimiters used in the scripts to isolate untrusted manuscript data from the generated report structure.
- Capability inventory: The skill utilizes local file system access to read inputs and write reports in Markdown, HTML, and CSV formats (e.g.,
09_qa/claim_audit.md). - Sanitization: The scripts perform regex-based extraction of numbers and patterns without sanitizing or escaping the input text before including it in the generated reports.
Audit Metadata