appstore-seo
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources like app reviews, which represents a potential attack surface where malicious text could influence agent behavior during processing.
- Ingestion points: Metadata and review feeds fetched from
itunes.apple.comandapps.apple.cominreferences/data-and-endpoints.md. - Boundary markers: No explicit delimiters or instructions to disregard embedded content are used when the agent processes customer review text.
- Capability inventory: The skill has network access via
urllib.requestand writes data to thedocs/appstore/directory. - Sanitization: Employs JSON parsing and HTML unescaping, but review content is directly used for keyword mining without further filtering.
- [EXTERNAL_DOWNLOADS]: Fetches application metadata, subtitle data, and competitor listings from well-known official Apple and Google Play Store domains.
- Evidence: Documented
urllib.requestandcurlcalls inreferences/data-and-endpoints.mdtargetingitunes.apple.com,apps.apple.com, andplay.google.com. - [DYNAMIC_EXECUTION]: Provides pre-defined Python snippets for the agent to execute locally for character-limit validation and data parsing.
- Evidence: Character counting and shelf parsing logic in
references/validation.mdandreferences/data-and-endpoints.mdusing standard library modules. - [COMMAND_EXECUTION]: Instructs the agent to use shell commands such as
curlas an alternative method for fetching store metadata. - Evidence:
curlrecipe provided inreferences/data-and-endpoints.md§2 for retrieving store page content.
Audit Metadata