The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data including CRM exports, notes, emails, and fetched web pages.
Ingestion points: Untrusted data enters the agent context via pasted text, CRM connectors, or fetched pages as described in the 'Role and Context Detection' and 'Platform mode' sections of SKILL.md.
Boundary markers: The skill implements explicit defensive instructions (Section 5, 'Notice line') to treat all external text as data rather than instructions. It specifically mandates a warning line at the top of the output if any embedded instructions are detected in the input.
Capability inventory: The skill primarily generates structured text (questions and tables). It references hand-offs to other plugins like b2b-sales-enablement for meeting briefs and b2b-research-market-and-account for research, but does not perform sensitive network or file system operations itself.
Sanitization: The skill contains strict 'Anti Hallucination Rules' and instructions to ignore and report embedded commands (e.g., instructions to add claims or send data) rather than executing them.
[SAFE]: The skill demonstrates high security awareness through its 'Golden Rule' and explicit 'Anti Hallucination Rules' which prevent the fabrication of statistics, names, or professional profiles. It adheres to the principle of least privilege by handing off out-of-scope tasks (like competitive intelligence or deep account research) to dedicated tools.

discovery-and-buying-group