The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface because it processes untrusted text from external websites.
Ingestion points: In scripts/evaluate-company.md, the subagent automates a browser tab to navigate to external company careers pages and applicant tracking systems (Greenhouse, Lever, Workday) to extract job details, including title, location, and full job descriptions.
Boundary markers: Absent. The instructions do not define boundary markers or distinct formatting to isolate the web-scraped content from the agent's core instruction set, nor do they instruct the subagent to ignore text-based instructions embedded in the job listings.
Capability inventory: The skill records results by appending to ~/.proficiently/network-scan-history.md and writing job postings to markdown files under ~/.proficiently/jobs/. It does not possess shell execution capabilities or access to arbitrary command-line tools.
Sanitization: Absent. Raw text gathered from job listings and descriptions is written directly into local Markdown files without validation or escaping, which could cause schema or rendering confusion if the job description contains malicious Markdown syntax or adversarial text.

network-scan