Hot-reload agent-authored skills and subagents #22

IgorWarzocha · 2026-01-15T22:34:17Z

Allow configs to be invalidated and reloaded while OC is running. This not only enables the users to create new agents/skills/commands and instantly use them, but it also allows the agents themselves to develop these skills or create new agents on the fly and instantly use them (Ralph Loops, anyone?).

Proposal:

Watch instance + config directories when hot-reload is enabled.
When agent/, skill/, or command/ markdown changes: invalidate cached state and reload immediately.
When custom tool .ts/.js changes: re-import with cache busting. (I believe this is a side effect that can be removed)
Emit config.updated, skill.updated, command.updated so TUI/Desktop refresh live.
Is it safe? Not necessarily, and is not for everyone. I have a functioning version of it, that is locked behind EXPERIMENTAL_HOT_RELOAD flag. Kept changes to minimum and preserved everything from upstream. https://github.com/IgorWarzocha/opencode/tree/agent-skills-commands-hot-reloading

It works, but there are some caveats - agents would need a create skill/command/agent tool for creating a markdown files that pass the parsing check (if it fails, nothing happens). This is probably the easiest implementation - let agent do the thing, and as a return of the tool call, give it a path to the file and ask it to expand the instructions set and leave the frontmatter alone.

Toast on reload would be a nice addition, but I'm not doing either until I know if there's actually a demand for this.

gemini-code-assist · 2026-01-15T22:34:53Z

Summary of Changes

Hello @IgorWarzocha, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces a significant quality-of-life improvement for developers and agents alike by implementing a hot-reloading capability. It allows for dynamic updates of agent configurations, skills, and commands without requiring a full application restart. This feature streamlines the development and iteration process, making it more fluid and responsive, particularly for agent-driven development workflows. The changes are carefully isolated behind an experimental flag to maintain stability.

Highlights

Hot-Reloading for Agents, Skills, and Commands: Introduced a new experimental feature that enables hot-reloading of agent-authored skills, subagents, and commands. This allows for instant updates without restarting the application when relevant configuration files or custom tools are modified.
File Watcher Integration: Implemented file watchers for configuration directories to detect changes in agent, skill, command markdown files, and custom tool JavaScript/TypeScript files. Upon detection, the relevant cached states are invalidated and reloaded.
Event-Driven UI Updates: New events (config.updated, command.updated, skill.updated) are now published when configurations change, allowing the TUI (Terminal User Interface) and other clients to refresh their live state dynamically.
Experimental Flag Control: The entire hot-reloading functionality is guarded by an OPENCODE_EXPERIMENTAL_HOT_RELOAD feature flag, ensuring it remains opt-in and does not affect stable environments by default.
Improved State Invalidation Mechanism: Enhanced the Instance and State management to support explicit invalidation of cached states, which is crucial for the hot-reloading logic.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request introduces an experimental hot-reloading feature for agents, skills, and commands, which is a great addition for faster development cycles. The implementation correctly uses file watchers and instance state invalidation. However, there are several areas for improvement, primarily concerning unhandled promises which could lead to silent failures, and duplicated code for path checking which impacts maintainability. I've also noted a missing watcher initialization for custom tools. Addressing these points will make the feature more robust and easier to maintain.

gemini-code-assist · 2026-01-15T22:38:01Z

packages/app/src/context/global-sync.tsx

+      case "config.updated": {
+        const sdk = createClient(directory)
+        Promise.all([
+          sdk.app.agents().then((x) => setStore("agent", x.data ?? [])),
+          sdk.command.list().then((x) => setStore("command", x.data ?? [])),
+          sdk.config.get().then((x) => setStore("config", x.data!)),
+        ])
+        break
+      }
+      case "command.updated": {
+        const sdk = createClient(directory)
+        sdk.command.list().then((x) => setStore("command", x.data ?? []))
+        break
+      }
+      case "skill.updated": {
+        const sdk = createClient(directory)
+        sdk.app.agents().then((x) => setStore("agent", x.data ?? []))
+        break
+      }


The promises returned by the SDK calls in the config.updated, command.updated, and skill.updated event handlers are not handled with .catch(). This can lead to unhandled promise rejections if an API call fails, causing silent failures in state updates. Additionally, in the config.updated handler, the non-null assertion x.data! is unsafe and could cause a runtime crash. Please add error handling to these promises and replace the non-null assertion with a safe access pattern.

gemini-code-assist · 2026-01-15T22:38:01Z

packages/app/src/context/global-sync.tsx

+        if (hasCommand) {
+          refresh.push(sdk.command.list().then((x) => setStore("command", x.data ?? [])))
+        }
+        Promise.all(refresh)


The Promise.all(refresh) call is not awaited and lacks a .catch() handler. If any of the promises in the refresh array reject (e.g., due to a network error), it will result in an unhandled promise rejection. This could cause the hot-reload to fail silently. It's safer to handle potential errors.

Promise.all(refresh).catch(err => console.error("Error during hot-reload refresh:", err));

gemini-code-assist · 2026-01-15T22:38:01Z

packages/opencode/src/agent/agent.ts

+      const configRoot = Global.Path.config.replaceAll("\\", "/")
+      const configDirs = await Config.directories()
+      const normalizedDirs = configDirs.map((dir) => dir.replaceAll("\\", "/"))
+      const looksLikeConfigDir =
+        filepath.includes("/.opencode/") ||
+        filepath.startsWith(".opencode/") ||
+        filepath.includes("/.config/opencode/") ||
+        filepath.startsWith(".config/opencode/")
+      const inConfigDir =
+        looksLikeConfigDir ||
+        filepath === configRoot ||
+        filepath.startsWith(configRoot + "/") ||
+        normalizedDirs.some((dir) => Filesystem.contains(dir, filepath) || filepath === dir)


This logic for determining if a file is within a configuration directory is duplicated in packages/opencode/src/command/index.ts, packages/opencode/src/config/config.ts, and packages/opencode/src/tool/registry.ts. To improve maintainability and reduce redundancy, consider extracting this into a shared helper function, for example in packages/opencode/src/config/config.ts.

gemini-code-assist · 2026-01-15T22:38:01Z

packages/opencode/src/cli/cmd/tui/context/sync.tsx

+          if (hasCommand) {
+            refresh.push(sdk.client.command.list().then((x) => setStore("command", reconcile(x.data ?? []))))
+          }
+          Promise.all(refresh)


The Promise.all(refresh) call is not awaited and lacks a .catch() handler. If any of the promises reject, it will result in an unhandled promise rejection, causing the TUI to not update on file changes without any visible error. Please handle potential errors.

Promise.all(refresh).catch(err => console.error("Error during TUI hot-reload refresh:", err));

gemini-code-assist · 2026-01-15T22:38:01Z

packages/opencode/src/cli/cmd/tui/context/sync.tsx

+        case "config.updated": {
+          setStore("config", reconcile(event.properties))
+          sdk.client.app.agents().then((x) => setStore("agent", reconcile(x.data ?? [])))
+          sdk.client.command.list().then((x) => setStore("command", reconcile(x.data ?? [])))
+          break
+        }
+
+        case "command.updated": {
+          sdk.client.command.list().then((x) => setStore("command", reconcile(x.data ?? [])))
+          break
+        }
+
+        case "skill.updated": {
+          sdk.client.app.agents().then((x) => setStore("agent", reconcile(x.data ?? [])))
+          break
+        }


In the new event handlers for config.updated, command.updated, and skill.updated, the promises returned by the SDK calls (e.g., sdk.client.command.list()) are not being handled with .catch(). This can lead to unhandled promise rejections if an API call fails, which might cause silent failures in the TUI state updates. Please add error handling to these promises.

gemini-code-assist · 2026-01-15T22:38:01Z

packages/opencode/src/project/bootstrap.ts

+  if (Flag.OPENCODE_EXPERIMENTAL_HOT_RELOAD()) {
+    Config.initWatcher()
+    Command.initWatcher()
+    Agent.initWatcher()
+    Skill.initWatcher()
+  }


The initWatcher functions for Config, Command, Agent, and Skill are correctly initialized here for hot-reloading. However, ToolRegistry.initWatcher() is also defined but is not being called. This means custom tools (.ts/.js files) will not be hot-reloaded as intended. Please add the call to ToolRegistry.initWatcher() here.

Suggested change

if (Flag.OPENCODE_EXPERIMENTAL_HOT_RELOAD()) {

Config.initWatcher()

Command.initWatcher()

Agent.initWatcher()

Skill.initWatcher()

}

if (Flag.OPENCODE_EXPERIMENTAL_HOT_RELOAD()) {

Config.initWatcher()

Command.initWatcher()

Agent.initWatcher()

Skill.initWatcher()

ToolRegistry.initWatcher()

}

ariane-emory · 2026-01-15T22:38:02Z

Neat! Big PR, but if it works well, I'd be into it. I'll try to give it a try this evening.

OpenCode Agent and others added 2 commits January 15, 2026 22:17

feat: hot-reload agent-authored skills and subagents

ce9e96b

Update node_modules hash (aarch64-linux)

d1fadc4

gemini-code-assist bot reviewed Jan 15, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Hot-reload agent-authored skills and subagents #22

Hot-reload agent-authored skills and subagents #22

Uh oh!

IgorWarzocha commented Jan 15, 2026

Uh oh!

gemini-code-assist bot commented Jan 15, 2026

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Uh oh!

gemini-code-assist bot Jan 15, 2026

Uh oh!

gemini-code-assist bot Jan 15, 2026

Uh oh!

gemini-code-assist bot Jan 15, 2026

Uh oh!

gemini-code-assist bot Jan 15, 2026

Uh oh!

gemini-code-assist bot Jan 15, 2026

Uh oh!

ariane-emory commented Jan 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Hot-reload agent-authored skills and subagents #22

Are you sure you want to change the base?

Hot-reload agent-authored skills and subagents #22

Uh oh!

Conversation

IgorWarzocha commented Jan 15, 2026

Uh oh!

gemini-code-assist bot commented Jan 15, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

ariane-emory commented Jan 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants