The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external job descriptions and job portal URLs which may contain adversarial instructions.
Ingestion points: External content enters the agent context through the auto-pipeline mode (JD text/URLs), scan mode (recruitment portals), and apply mode (application forms).
Boundary markers: The instructions include an output language directive that mandates consistency, but the skill lacks explicit delimiters or specific instructions to ignore embedded commands within the ingested job descriptions.
Capability inventory: The agent has the capability to write local files (CVs, trackers), spawn subagents for processing, and perform web automation tasks via tools like Playwright.
Sanitization: No explicit sanitization or filtering of external content is described before it is processed by the agent.
[DYNAMIC_EXECUTION]: The update mode allows the agent to modify its own system and configuration files, representing a self-modification capability.
[COMMAND_EXECUTION]: The skill relies on the execution of external commands for browser automation (Playwright) and document generation (LaTeX).

career-ops