The Agent Skills Directory

[SAFE]: The skill consists entirely of structured markdown instructions and templates for data organization. It contains no executable scripts (Python, JavaScript, shell), preventing risks related to command execution or remote code vulnerabilities.
[DATA_EXFILTRATION]: While the skill involves gathering professional contact information (PII) such as email patterns and leadership names, it is restricted to publicly available professional data for sales intelligence. It includes specific guardrails: 'Do not collect personal information beyond what is publicly available and professionally relevant' and 'This intelligence is for legitimate sales and marketing use — never for harassment, spam, or deceptive practices.'
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze data from external websites, job postings, and news articles. This constitutes a standard ingestion surface for indirect prompt injection.
Ingestion points: Target company domains, job postings, LinkedIn profiles, and news articles (Phase 1, 2, and 3).
Boundary markers: The skill uses structured YAML templates for all phases, providing a schema that naturally delimits data from instructions.
Capability inventory: None. The skill itself has no capability to write files, execute subprocesses, or perform network operations directly.
Sanitization: Not explicitly implemented, but the use of strict output schemas reduces the likelihood of an agent interpreting data as instructions.

customer-intel