appstore-review
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user project files (source code, configuration) and dynamic content fetched from Apple's developer site. This creates a vulnerability surface where instructions embedded in the audited files or the fetched guidelines could attempt to influence the agent's behavior.
- Ingestion points: Apple's guidelines and news text (
guidelines.txt,news.md) and project source code/metadata located in the local file system. - Boundary markers: The skill uses specific prompt sections (e.g., 'LIVE SOURCE TEXT') and explicit instructions to the agent to rely solely on the fetched text and return structured output without commentary.
- Capability inventory: The skill reads local files, identifies project structure via
gitorfind, and executes a bundled Python script that usescurlfor network requests. - Sanitization: The
fetch_sources.pyscript strips HTML tags during conversion to plain text, but the skill lacks specific sanitization for natural language instructions contained within the text data. - [EXTERNAL_DOWNLOADS]: The skill fetches the latest App Store Review Guidelines and Developer News from Apple's official domains.
- Fetches content from
developer.apple.com/app-store/review/guidelines/anddeveloper.apple.com/news/. - The data is used as local context for the audit process.
- [COMMAND_EXECUTION]: The skill invokes a bundled Python utility to retrieve and process external documentation.
- Evidence: The skill runs
python3 <skill-dir>/scripts/fetch_sources.pyat the start of the workflow and provides it to lane agents for rule lookups. - Implementation: The script uses
subprocess.runwith a list of arguments to callcurl, targeting hardcoded official Apple URLs.
Audit Metadata