video-digest-site
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown and JSON data which could contain malicious instructions designed to influence an AI agent that subsequently processes the generated corpus.
- Ingestion points: The scripts
scripts/vd_corpus.py(line 33) andscripts/vd_site.py(lines 88 and 114) ingest content from markdown files and JSON data located in the project'sgroups/directory. - Boundary markers: The generated
narrative_corpus.mddoes not utilize boundary markers or clear delimiters to separate ingested content from the agent's context, potentially allowing embedded instructions to be interpreted as system commands by downstream models. - Capability inventory: The skill has the capability to write files to the local filesystem (
out.write_text) and perform image processing via Pillow. - Sanitization: While
scripts/vd_site.pyuses basic HTML escaping for the website output,scripts/vd_corpus.pydoes not sanitize or wrap the text content when generating the synthesis essay corpus.
Audit Metadata