The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted external content (transcripts, articles, media links) to generate derivatives, creating a surface for indirect prompt injection. * Ingestion points: Pillar source content (Phase 1) and external transcript/video material (Phase 2). * Boundary markers: Absent; no specific delimiters or instruction-isolation markers are defined for source content. * Capability inventory: File-read and file-write access to the project root and workspace directory. * Sanitization: None explicitly performed on external inputs.
[COMMAND_EXECUTION]: The skill references local scripts within the Kai installation's internal directory. * Evidence: The instructions include a safety check requiring the agent to report missing scripts rather than attempting to execute unknown or fabricated commands.
[DATA_EXFILTRATION]: The skill automates discovery of local codebase settings, including analytics and email configurations, to generate a product profile. * Scope: File access is restricted to the local environment for generating the MARKETING.md file; no network exfiltration or credential theft patterns were detected.

kai-repurpose