The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability during the identity import process.
Ingestion points: aieos-import.js reads untrusted data from external JSON files (e.g., default.aieos.json).
Boundary markers: The generated markdown files (SOUL.md, IDENTITY.md, AGENTS.md) do not include delimiters or specific instructions for the agent to ignore potentially malicious content within imported fields.
Capability inventory: The skill includes file writing operations (fs.writeFileSync in aieos-import.js) and subprocess execution (child_process.execSync in test.js).
Sanitization: There is a lack of sanitization for user-controlled strings (such as core_values, catchphrases, and style_descriptors) before they are written into the agent's behavior configuration files.
[COMMAND_EXECUTION]: The test.js utility utilizes child_process.execSync to execute local Node.js scripts. This is a standard testing pattern but involves the execution of system-level commands based on path variables.

aieos-identity