app-store-optimization
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [PROMPT_INJECTION]: The
SKILL.mdfile contains instructions in the 'Communication' section that attempt to dictate the agent's output format ('Bottom Line → What → Why → How to Act') and self-verification process, which constitutes an attempt to override default agent behavior. - [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data such as user-provided app store reviews and competitor descriptions in
review_analyzer.pyandcompetitor_analyzer.py. This exposes the agent to indirect prompt injection risks if malicious instructions are embedded within the text data being analyzed. Mandatory Evidence Chain: (1) Ingestion points:ReviewAnalyzer.analyze_sentiment(file:scripts/review_analyzer.py),CompetitorAnalyzer.analyze_competitor(file:scripts/competitor_analyzer.py). (2) Boundary markers: Absent. (3) Capability inventory: The Python scripts themselves have no dangerous capabilities, but the agent processing the output may follow instructions in the text. (4) Sanitization: Basic punctuation removal via regex. - [METADATA_POISONING]: The
README.mdfile incorrectly identifies the author as 'Claude Skills Factory', creating a discrepancy with the actual author attribution. This constitutes misleading metadata.
Audit Metadata