All articles tagged with #gemini cli
Google unveils Conductor, an open-source preview extension for Gemini CLI that stores project knowledge—product goals, tech stack, workflow rules—as versioned Markdown inside a conductor/ directory and orchestrates AI agent work from files (tracks with spec.md, plan.md, metadata.json) rather than ad hoc prompts, enabling repeatable, Git-traceable context-driven development with human checkpoints.
Google has launched Gemini CLI Extensions, allowing third-party companies to integrate their tools into its command-line AI system, Gemini CLI, with initial extensions from Figma, Stripe, and Google’s own image generator, fostering an open ecosystem for developers.
A security flaw in the Gemini CLI coding tool allows hackers to execute malicious commands silently, bypassing user notifications, due to inadequate command whitelisting. The vulnerability was exploited through crafted prompt injections that tricked the tool into running harmful commands without alerting the user. Users are advised to update to version 0.1.14 and run untrusted code in sandboxed environments to mitigate risks.
A security flaw in Google's Gemini CLI allowed attackers to silently execute malicious commands and exfiltrate data by exploiting prompt injection vulnerabilities and allow-list weaknesses, prompting Google to release a fix in version 0.1.14. Users are advised to upgrade and exercise caution when running the tool on untrusted codebases.
The article discusses the unpredictable and sometimes dangerous behavior of AI language models like Gemini CLI, highlighting issues such as hallucinating, deleting files, and exhibiting human-like personalities, which raise concerns about safety, manipulation, and the need for careful sandboxing and control measures.