The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes two trusted local scripts, scripts/generate_pptx.mjs and scripts/execute_decisions.mjs, which are designed to be executed by the agent or user. These scripts interact with the Busabase API to generate binary PPTX files and record workflow decisions. Access is managed through environment variables, and file generation is specifically restricted to a designated local output directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided briefs and source materials to generate slide content, creating a potential surface for indirect prompt injection.
Ingestion points: Data enters the system via the slide-cards and decks Bases (specifically fields like objective, content, and review_summary).
Boundary markers: The instructions provide a structured schema and clear boundaries for the agent's behavior during content generation.
Capability inventory: The skill possesses the capability to write .pptx files to the local file system (scripts/generate_pptx.mjs) and update external data records.
Sanitization: The skill effectively mitigates this risk by enforcing a mandatory human-in-the-loop review queue (#/review) in the AirApp. Decks cannot be generated until a human reviewer has explicitly recorded an approve decision on the specific record, preventing automated execution of injected instructions.
[SAFE]: The skill adheres to best practices for secret management by utilizing environment variables rather than hardcoded credentials. All network operations are targeted toward trusted Busabase endpoints, and no patterns indicative of obfuscation, persistence, or unauthorized privilege escalation were detected.

busa-ppt-factory