Skip to content

v1.1.54: Skill URL Discovery, More Stable ACP File Writing, Clearer Model Errors ​

Release Date: February 11, 2026


One-Sentence Summary ​

This update completes the "most common pitfalls in major versions": Skills can be auto-discovered via URL, ACP file writing automatically creates new files, model selection errors provide clear events and error types, and desktop/TUI fixed a batch of high-frequency issues.


New Feature: Skill Auto-Discovery via URL ​

What's Updated? ​

OpenCode added Skill URL discovery capability based on well-known RFC, making subsequent Skill integration via URL smoother.

What's the Use for You? ​

Previously you needed to manually find Skill entry points; now you can go through the discovery process directly from standard URLs, especially suitable for teams to uniformly distribute Skills.

How to Use? ​

👉 Full Tutorial: Skills Advanced

👉 Official Docs: Agents

Quick Experience:

  1. First confirm your Skill source address is publicly accessible.
  2. Configure and trigger loading according to existing Skill process in your project.
  3. Observe whether the corresponding Skill is auto-discovered and loaded via URL.

New Feature: ACP Auto-Creates Non-Existent Files When Writing ​

What's Updated? ​

When writing files via ACP channel, if the target file doesn't exist, OpenCode now automatically creates the file before writing.

What's the Use for You? ​

When you execute write operations in IDE + ACP scenarios, it no longer fails directly because "file hasn't been created yet", making automation scripts more stable.

How to Use? ​

👉 Full Tutorial: ACP Protocol

👉 Official Docs: ACP Support

Quick Experience:

  1. Start ACP according to tutorial (opencode acp).
  2. Let Agent write to a non-existent path.
  3. Confirm file is created and written successfully.

New Feature: Enhanced Model Selection Error Events and Error Types ​

What's Updated? ​

Added session.error event for invalid model selection; also added ContextOverflowError, and improved model variant availability check logic.

What's the Use for You? ​

Troubleshooting model issues is no longer "silent failure" or vague errors. Logs and events will tell you faster what went wrong.

How to Use? ​

👉 Full Tutorial: Config Basics

👉 Extended Tutorial: Thinking Depth Configuration

👉 Official Docs: Models

Quick Experience:

  1. Deliberately fill in a non-existent model ID in config.
  2. Start a session and observe event or log output.
  3. Change back to an available model, verify session returns to normal.

Bug Fix: Multiple Stability Fixes for Desktop / TUI / Core ​

Previous Issue ​

Including Homebrew upgrade occasionally requiring multiple runs, app terminal replay anomalies, unstable behavior after workspace reset, etc.

Current Status ​

These issues are centrally fixed in v1.1.54, with significantly improved stability for desktop, TUI, and core workflows.


Should You Upgrade? ​

SituationRecommendation
You're using ACP or IDE integration✅ Recommend immediate upgrade
You frequently switch models/variants✅ Recommend immediate upgrade
You mainly use desktop✅ Recommend upgrade
Everything works fineCan wait for next time

Upgrade Command:

bash
opencode upgrade

Technical Details (Skippable) ​

Click to expand full update list

Core (Excerpt) ​

  • Publish session.error event for invalid model selection
  • Create file if it doesn't exist when writing via ACP
  • Add new ContextOverflowError type
  • Add skill discovery from URLs via well-known RFC
  • Parse mid-stream OpenAI response errors to prevent infinite retries
  • Resolve Homebrew upgrade requiring multiple runs

TUI (Excerpt) ​

  • Use sender color for queued messages in TUI
  • Add --fork flag to duplicate sessions before continuing
  • Improve skills dialog readability

Desktop (Excerpt) ​

  • Fix terminal replay issues in app
  • Fix workspace reset and session navigation issues
  • Improve prompt input, sidebar, and file tree interactions
  • Add native Wayland toggle and more menu bar capabilities

Official Release Notes:


← Back to Changelog