save
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (YouTube transcripts, web pages, and documents) to generate Obsidian notes and metadata. This creates an attack surface where instructions embedded in the source material could potentially influence the agent's behavior during restructuring or metadata generation steps.\n
- Ingestion points: External content is fetched via
extract-defuddle.sh,extract-web.sh, andextract-doc.shfrom user-supplied URLs and file paths.\n - Boundary markers: The skill does not explicitly use delimiters or "ignore embedded instructions" warnings when passing extracted content to sub-tasks for restructuring or translation.\n
- Capability inventory: The skill has permissions to write files to the local repository (
01-inbox/), execute shell scripts, and modify git state (branch renaming).\n - Sanitization: Branch names and filenames are sanitized (converted to lowercase, special characters removed) before use.
jqis used for safe JSON construction in all extraction scripts.\n- [EXTERNAL_DOWNLOADS]: The skill dynamically downloads and executes themarkitdownpackage from PyPI during the document processing phase.\n - Evidence:
scripts/extract-doc.shusesuvx --from \"markitdown[all]\" markitdown \"$FILE_PATH\"to convert documents to markdown.\n- [COMMAND_EXECUTION]: The skill performs several shell-based filesystem and git operations based on the processed content.\n - Evidence: The orchestrator and scripts use commands including
git branch -m,rm -rf,cat, andgrepfor context management and repository organization.
Audit Metadata