OpenCode v1.18.2 improves desktop navigation and session visibility while introducing bounded subagent recursion and resilient Windows directory creation.

// 1h agoPRODUCT UPDATE

OpenCode v1.18.2 improves desktop navigation and session visibility while introducing bounded subagent recursion and resilient Windows directory creation.

OpenCode has released version 1.18.2, featuring enhancements to desktop navigation and session visibility, including the addition of Cmd/Ctrl+N as an alternate shortcut for opening a new tab. This update introduces configurable and bounded subagent recursion to prevent runaway loops, defaults Meta model settings to "xhigh" reasoning, and improves the reliability of directory creation on Windows systems.

// ANALYSIS

Bounding subagent recursion is a vital safeguard against runaway API costs, making this update a necessary maturity step for autonomous developer tools.

* Configurable recursion limits provide necessary control over autonomous agent loops and API spend.

* Shifting Meta models to default to 'xhigh' reasoning points to a preference for deeper reasoning steps in code generation tasks.

* The Cmd/Ctrl+N shortcut and improved session visibility refine the app's user experience to match developer expectations.

* Windows-specific filesystem improvements ensure the agent remains robust across different development environments.

// TAGS

opencodeai-codingcoding-agentagentdevtoolframework

DISCOVERED

1h ago

2026-07-15

PUBLISHED

2h ago

2026-07-15

RELEVANCE

7/ 10

AUTHOR

OpenCodeLog

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

OPEN SOURCE32m ago

Netlify drops context-and-tools for autonomous AI agent deployments

A new tip highlights that AI coding tools read AGENTS.md files for context. To capitalize on this, Netlify has published a public skills repository containing an AGENTS.md file that equips AI agents with the knowledge needed to handle netlify.toml configurations, generate previews, and execute production deployments autonomously simply by dropping it into a project.

NEWS40m ago

GPT-Red adversarially trains GPT-5.6 against prompt injections

A recent update highlights the high return on investment of self-improving AI models, specifically noting how GPT-Red was used to adversarially train GPT-5.6. This training approach has reportedly made the model much more robust against prompt injection vulnerabilities, showcasing a vital method for improving AI security.

UPDATE47m ago

Mintlify MCP turns agents into doc reviewers

Mintlify has launched a new Model Context Protocol (MCP) feedback tool that turns AI agents into active documentation reviewers. By allowing agents to flag gaps and confusing sections as they read, the tool provides documentation creators with precise, actionable insights to improve their content based on real AI interactions.