The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes browser automation tools (tabs_create_mcp, form_input, computer) to navigate job boards and interact with form elements. These actions are restricted to the primary purpose of the skill and include a mandatory user confirmation step before the final 'Submit' action is performed.
[DATA_EXPOSURE]: The skill processes and manages sensitive personal identifiable information (PII), such as legal names, contact details, and resumes, stored in a local .proficiently/ directory. Access is limited to the local filesystem and the target job application domains (e.g., Lever, Greenhouse, Workday) provided in the workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external job postings via read_page and get_page_text. While this presents a surface for indirect prompt injection, the risk is mitigated by the implementation of a constrained subagent (scripts/fill-page.md) that follows a pre-approved mapping and is instructed not to make autonomous decisions based on page content.
Ingestion points: read_page and get_page_text are used in SKILL.md to parse external job postings.
Boundary markers: None explicitly used to wrap external content.
Capability inventory: form_input, computer (click/type), and local file writes (DATA_DIR).
Sanitization: No explicit sanitization or filtering of external text is performed before processing.

apply