The Agent Skills Directory

[SAFE]: The skill is entirely composed of documentation and templates for social media content. No executable scripts, network requests, or credential-harvesting patterns were identified.
[SAFE]: The 'Operating contract' in the main skill file includes a critical safety constraint requiring user approval before the agent can post, schedule, or reply publicly, which prevents unauthorized autonomous actions.
[INDIRECT_PROMPT_INJECTION]: The skill defines workflows for processing untrusted data from external social media platforms (listening briefs) and local context files.
Ingestion points: Monitored social media accounts, 'listening' targets, and the .agents/marketing-context.md file.
Boundary markers: The skill identifies specific files as 'truth' constraints but does not define explicit delimiters for content retrieved from social media.
Capability inventory: Instructions allow for the generation and scheduling of public content.
Sanitization: No specific data sanitization steps are defined for external content.
Mitigation: The risk is addressed by the mandatory requirement for human review and approval of all generated assets before publication.

organic-social