The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the oma slide CLI for deterministic operations such as scaffolding workspaces, performing geometric validation of slides via Puppeteer, and bundling assets. These subprocess calls are standard for the skill's primary functionality and use localized paths.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from external sources, specifically font assets from well-known CDNs (Google Fonts, Bunny Fonts, Adobe Typekit, and jsDelivr) and style templates from a public GitHub repository. These are documented as functional dependencies, and the protocol includes specific instructions to treat remote style data as untrusted and to provide fallbacks in case of fetch failures.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes external style templates and user-provided assets (PPTX files, Canva designs, and images). It includes a mandatory evidence chain to mitigate this risk:
Ingestion points: Remote design.md fetching in Phase 2 and file imports in Phase 0.
Boundary markers: The skill explicitly instructs the agent to treat fetched data as 'untrusted' and to log the source of the content.
Capability inventory: Capabilities include writing slide fragments to the local file system and executing CLI tools for rendering and exporting.
Sanitization: The generation protocol mandates sanitization of external style data and includes automated recovery logic for malformed inputs.
[COMMAND_EXECUTION]: To support Canva integration, the skill manages MCP configuration by offering to write to project-level and user-global configuration files (e.g., ~/.gemini/antigravity-cli/mcp_config.json). This sensitive file system operation is strictly gated by a requirement for explicit user approval before any changes are committed.

oma-slide