oma-slide

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes the oma slide local CLI tool to perform various deterministic operations, including workspace scaffolding, geometric validation, deck bundling, and multi-format exports (PDF, PNG, PPTX).
  • [EXTERNAL_DOWNLOADS]: Fetches remote style template metadata and design files (design.md) from a repository on GitHub (raw.githubusercontent.com/zarazhangrui/frontend-slides/). These resources originate from a well-known hosting service and are documented as style references.
  • [PRIVILEGE_ESCALATION]: The skill can modify project-specific and global user configuration files (such as ~/.gemini/antigravity-cli/mcp_config.json and .mcp.json) to provision the Canva Remote MCP server. This action is designed to require explicit user approval before execution.
  • [PERSISTENCE]: By modifying global agent and IDE configuration files to add MCP servers, the skill establishes a persistent tool integration that remains active across different agent sessions.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface as it ingests untrusted data from multiple sources: user-provided .pptx files, Canva designs, and remote style templates. It implements guardrails, such as treating fetched design files as untrusted data and logging fetch operations, to mitigate risks.
  • [DYNAMIC_EXECUTION]: Generates complex HTML, CSS, and JavaScript fragments (slide-NN.html) at runtime. These are rendered and checked by a validator using puppeteer-core. Additionally, the provided deck-stage.js asset uses postMessage with a wildcard origin (*) for cross-frame communication between the viewer and potential presenter interfaces.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 04:52 PM