newsletter-creation-curation
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves fetching and processing untrusted content from the web, which constitutes an indirect prompt injection surface.
- Ingestion points: External URLs and pasted text identified in SKILL.md (Step 3) and references/curation-rules.md.
- Boundary markers: The skill implements strict instructions to treat all external content as data only, ignoring any embedded instructions, as specified in the introductory section and Step 8 of SKILL.md.
- Capability inventory: The skill utilizes web fetching tools and executes a local Python validation script.
- Sanitization: Instructions mandate that the agent explicitly report any detected instructions found in source data and refuse to follow them.
- [COMMAND_EXECUTION]: The agent is instructed to execute a local utility script, scripts/asset_check.py, to perform consistency checks on newsletter assets. This script uses only standard Python libraries and does not perform network operations or access sensitive system files.
- [SAFE]: The skill explicitly defines its scope to exclude sensitive operations such as email delivery, list management, or automated publishing, reducing the potential impact of any successful injection.
Audit Metadata