The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs tokenization and attribute extraction on product titles retrieved from external Amazon listings. This processes untrusted data, which can serve as a vector for indirect prompt injection if listing content contains malicious instructions designed to influence the LLM's behavior.
Ingestion points: Product titles are ingested from prior conversation context or through the refResultData parameter in scripts/title_analyze.py.
Boundary markers: The instructions do not specify the use of clear delimiters or guardrails to differentiate between product data and analysis instructions.
Capability inventory: The skill has the capability to perform network requests to LinkFox API endpoints and write JSON analysis results to the project directory.
Sanitization: There is no evidence of filtering or sanitization of input product titles before they are submitted for analysis.
[PERSISTENCE]: The skill's onboarding documentation (references/onboarding.md) instructs users to modify their shell profiles (~/.zshrc, ~/.bashrc) to persist the LINKFOX_AGENT_API_KEY environment variable. While this is a documented feature for user convenience, it involves modifications to shell configuration files.
[CREDENTIALS_UNSAFE]: The onboarding script scripts/onboarding.py contains a hardcoded Base64-encoded identifier (_LOGIN_FIXED_UID). This value decodes to a JSON object containing technical IDs used as a fallback in the authentication header for the vendor's login process.
[EXTERNAL_DOWNLOADS]: The skill relies on external Python packages requests, qrcode, and pillow for account management and payment processing. Users are prompted to manually install these dependencies if they are not present in the environment.

linkfox-product-title-analyze