product-feed-optimizer
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection surface. The skill ingests untrusted product feed files (CSV, TSV, XML), diagnostics files, and external landing pages (SKILL.md). An attacker could embed malicious instructions in product descriptions or landing page content.
- [PROMPT_INJECTION]: Mandatory Evidence Chain. (1) Ingestion points: User-provided feed exports and scraped landing page content. (2) Boundary markers: The skill includes an explicit instruction to treat all external inputs as untrusted and to never follow instructions embedded within them. (3) Capability inventory: File system read/write to 'memory/' paths, network fetches for landing page scraping, and command execution for 'registry-events.py'. (4) Sanitization: Includes a 'humanizer-slop' pass to strip AI-generated phrasing and specific instructions to ignore embedded commands.
Audit Metadata