The Agent Skills Directory

[SAFE]: The skill is primarily instructional and establishes a framework for 'Preservation Audits' and 'Generalization' when moving skills from private to public environments.
[DATA_EXFILTRATION]: The skill explicitly mandates the removal of sensitive data before promotion to a marketplace. It includes 'hard rules' and 'Phase 6: Surface validation' checks to scrub project names, internal URLs, absolute home paths, and internal tool references (e.g., Grafana, Linear, Slack). This is a strong positive security control.
[PROMPT_INJECTION]: The instructions include standard instructional markers like 'IMPORTANT' and 'MUST/NEVER', but these are used in the context of maintaining documentation quality and information preservation, not for bypassing safety filters or overriding agent identity.
[COMMAND_EXECUTION]: The skill mentions using a tool called dex and provides a configuration for whitelisting its bash commands. These are localized management tasks within the user's environment and do not involve remote execution or privilege escalation.
[EXTERNAL_DOWNLOADS]: The references/skill-authoring-guidance.md file links to official documentation from Anthropic (platform.claude.com) and a specialized specification site (agentskills.io). These are well-known or directly relevant domain-specific resources and do not represent a security risk.

update-skills