daily-papers-notes
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of paper titles and method names extracted from external academic sources.
- Ingestion points: The skill reads from
/tmp/daily_papers_enriched.jsonand Markdown recommendation files ({DAILY_PAPERS_PATH}/YYYY-MM-DD-论文推荐.md) to extract paper information. - Boundary markers: The instructions lack explicit delimiters or warnings to the agent to ignore instructions that might be embedded within the ingested paper metadata (e.g., in a paper title or abstract).
- Capability inventory: The skill can modify Markdown files using the
backfill_links.pyscript, execute local Python tools (generate_concept_mocs.py,generate_paper_mocs.py), and perform Git operations (git commit,git push). - Sanitization: The
backfill_links.pyscript appropriately usesre.escape()when building regex patterns from extracted method names, which provides protection against malformed strings during the backfilling process.
Audit Metadata