OpenCode hardens Plan Mode, tool schemas

// 1h agoOPENSOURCE RELEASE

OpenCode hardens Plan Mode, tool schemas

OpenCode v1.14.46 is a security- and reliability-focused release for the terminal AI coding agent. The update makes Plan Mode subagents inherit deny rules, reduces drift in HTTP/OpenAPI query schemas, tolerates malformed MCP tool discovery output schemas, and loads legacy session data more defensively. In practice, this is the kind of release that lowers the odds of subtle agent-policy bypasses and flaky integrations.

// ANALYSIS

This is a strong maintenance release because it targets the failure modes that matter most in an agent runtime: permission leakage, schema drift, and brittle backward compatibility.

–Plan Mode subagents now inherit deny rules, which closes a meaningful guardrail gap for delegated agent work.
–The HTTP/OpenAPI schema fix should reduce tool-call drift, especially in integrations that rely on generated query params.
–MCP discovery becoming tolerant of bad output schemas is a pragmatic interoperability win; it trades strictness for survivability.
–Defensive loading for legacy session data suggests the team is prioritizing upgrade safety over perfect historical fidelity.
–Net: not flashy, but the right kind of release for a product that can write files and orchestrate tools.

// TAGS

opencodereleaseplan-modesecuritymcpopenapiagents

DISCOVERED

1h ago

2026-05-10

PUBLISHED

1h ago

2026-05-10

RELEVANCE

9/ 10

AUTHOR

OpenCodeLog

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

OPEN SOURCE17m ago

HeyGen open-sources HyperFrames under Apache 2.0

HeyGen is opening HyperFrames as an Apache 2.0 project on GitHub, positioning it as an agent-native HTML-to-video framework for creating and rendering video compositions from HTML, CSS, and JavaScript. The announcement also highlights a new /contribute-catalog skill intended to lower the friction for community contributions, so users can focus on building creative video experiences while the tooling handles the repetitive work.

TUTORIAL23m ago

Anthropic workshop reveals Claude prompting playbook

Anthropic's Applied AI team published "The prompting playbook," a recorded workshop from Code w/ Claude 2026 that shows how they build and maintain production prompts across model and architecture changes. The session focuses on iterative prompt debugging, XML structure, and explicit task ordering rather than prompt mysticism.

MODEL29m ago

Genesis AI unveils GENE-26.5 robotics brain

Genesis AI says GENE-26.5 is its first robotics foundation model, aimed at human-level dexterous manipulation across cooking, lab work, wire harnessing, and other long-horizon tasks. The release is bundled with a human-scale robotic hand, a glove-based data engine, and simulation stack meant to remove the usual robotics data bottleneck.