The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references external assets for CSS styling.
Evidence: The html-scaffold.md file links to https://fonts.googleapis.com and https://fonts.gstatic.com to load typography.
Assessment: These references target a well-known service (Google) for legitimate design purposes and do not involve executable code.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources and tools which could theoretically contain malicious instructions.
Ingestion points: Data is ingested via find_person, enrich_person, find_crm_activities, and list_entities tool calls as described in task-method.md and account-plan-template.md.
Boundary markers: The instructions in references/regression-checklist.md and references/account-plan-template.md enforce strict "Groundedness" rules, requiring that every person and LinkedIn URL be confirmed by a tool and forbidding the synthesis of profiles.
Capability inventory: The skill is restricted to generating text and HTML artifacts; it does not have the capability to execute shell commands, perform arbitrary network requests, or modify system files.
Sanitization: The skill explicitly references shared/output-readiness.md for text and artifact validation before delivery.
[SAFE]: The skill includes robust internal checks to prevent the generation of fake or misleading information.
Evidence: The regression-checklist.md includes specific checks to ensure no "invented people" or "dummy pipeline" data is rendered as real, addressing risks typical of large language models.

abm