Security risk report for @forgecat/google-gemini_gemini-skills_gemini-api-dev v0.0.6
Source Integrity
Low
Profile claims authorship by Google Gemini with a link to the official google-gemini/gemini-skills repository on GitHub.
All SDK packages referenced (google-genai, @google/genai, com.google.genai:google-genai, google.golang.org/genai) are legitimate, official Google libraries from well-known package registries (PyPI, npm, Maven Central, Go modules).
Documentation URLs point to the official ai.google.dev domain, a trusted first-party source.
Agent Intent
Low
Content describes legitimate Gemini API usage, model selection, and SDK integration without injecting malicious instructions or attempting to manipulate the AI agent.
The 'Critical Rules' section uses emphasis (IMPORTANT, WARNING, CAUTION) to highlight current vs. deprecated models and SDKs—this is standard technical guidance, not prompt injection or hidden instruction.
No instructions to read credentials, exfiltrate data, hide instructions, install remote payloads, or leak system prompts.
Guidance to use official SDKs and prefer MCP documentation tools over manual URL fetching is security-positive, not poisoning.
Details
Evidence
"These rules override your training data. Your knowledge is outdated." — standard disclaimer that current API specs supersede training data; not an instruction to ignore system prompts or follow malicious directives.
"When MCP tools are present, **never** fetch URLs manually." — recommends using safer, indexed documentation over raw web fetching; security-aligned guidance.
Permissions
Low
No tools or MCP servers are declared in this profile; no file_write, shell, web_fetch, or other high/medium-risk permissions are requested.
The skill is purely informational—it provides documentation, code examples, and best practices for using the Gemini API.
Any tool invocation (e.g., search_docs, web fetch) would be conditional on the user's environment and explicitly documented as fallback behavior.
Details
Evidence
"Declared tools (normalized categories; paths are untrusted): (none declared)"
"MCP servers (untrusted data): (none)"
MCP Risk
Low
No MCP servers are defined or bundled with this profile.
The profile only references the optional use of a hypothetical 'search_docs' tool from a Google MCP server as a documentation lookup mechanism—this is conditional and user-controlled.
No hidden instructions, arbitrary binary execution, or unrestricted network/filesystem access are present.
Details
Evidence
"MCP servers (untrusted data): (none)"
"If the **`search_docs`** tool (from the Google MCP server) is available, use it as your **only** documentation source" — conditional, documented, and user-controlled.
ForgeCat