The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to install a core dependency using npx skills add suzheng/k2f --skill k2f. This command fetches a resource from the skill author's own repository/ecosystem to provide the underlying formatting engine.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and synthesize external user data, such as drafts, notes, and PDFs. This creates a surface for indirect prompt injection if the source materials contain malicious instructions.
Ingestion points: User-provided inputs including drafts, PDFs, and field notes are processed in Step 1, Sub-step 3 (Source synthesis).
Boundary markers: The instructions include an "Honesty gate" which explicitly directs the agent to treat user materials as data to be synthesized rather than as instructions to override the skill's scope or rules.
Capability inventory: The skill generates markdown blueprints (content_and_design.md) and utilizes the k2f skill to create and export .K2F package files (PDF, DOCX).
Sanitization: The skill mandates deduplication, clustering by theme rather than filename, and flagging conflicting interpretations, which helps the agent maintain an analytical distance from the input content.

k2f-thesis