The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded cryptographic salts used for session state protection were found in the application configuration files. While these are part of the application definition metadata, they represent sensitive configuration data that should be uniquely generated and managed as secrets in deployed environments.
Evidence: checksumSalt: 40ACB0916413E0771D12A103716D68895FC91CBBB406CAFA608646F1C885FCA0 in assets/example-1/application.apx.
Evidence: checksumSalt: B98FE33A52A9044B17595883F386B708F82408C0D23D099041CCF5BBF896CBB8 in assets/example-2/application.apx.
Evidence: checksumSalt: BE8E26C19CF3511FBFEB1A6F9F79DA741A84B7ED70299369E025F6E56E2FFF65 in assets/example-0/application.apx.
[INDIRECT_PROMPT_INJECTION]: The example applications include features that ingest and process untrusted external data, creating an attack surface for indirect prompt injection when these applications are viewed or manipulated by an AI agent.
Ingestion points: Feedback submission forms (e.g., P10050_FEEDBACK in assets/example-1/pages/p10050-feedback.apx) and data loading via file upload (e.g., P22_FILE in assets/example-2/pages/p00022-khaeld.apx).
Boundary markers: Data ingestion components lack specific delimiters or instructions to ignore embedded malicious prompts during processing.
Capability inventory: The applications use platform APIs such as apex_data_parser.parse for processing external file content and APEX_UTIL.SUBMIT_FEEDBACK for data storage.
Sanitization: While some components explicitly use apex_escape.html in display logic, the overall security posture relies on the underlying platform's default escaping mechanisms, which may be bypassed if specific item settings are configured (e.g., format: htmlUnsafe used for P10054_RATING_ICON in assets/example-1/pages/p10054-feedback1.apx).

apexlang-example-applications