local: plugin init, reasoning_content stripping, debug logging, pvrdrxtd submodule

- Strip reasoning_content/reasoning_details from messages for providers that
  don't support interleaved reasoning (e.g. Cerebras)
- Add plugin.init() calls in Agent.state and Command layer so plugins load
  before skills/commands are built
- Add LLM param debug logging for model introspection
- Add pvrdrxtd submodule and .gitignore entries for local paths

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: rustybret

.gitignore

@@ -30,3 +30,7 @@ UPCOMING_CHANGELOG.md
 logs/
 *.bun-build
 tsconfig.tsbuildinfo
+
+/.opencode/plugins
+/.sisyphus/run-continuation
+/.teams/runtime

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "submodules/pvrdrxtd"]
+	path = submodules/pvrdrxtd
+	url = https://github.com/rustybret/pvrdrxtd.git

packages/opencode/src/agent/agent.ts

@@ -88,6 +88,7 @@ export const layer = Layer.effect(
 
     const state = yield* InstanceState.make<State>(
       Effect.fn("Agent.state")(function* (ctx) {
+        yield* plugin.init()
         const cfg = yield* config.get()
         const skillDirs = yield* skill.dirs()
         const whitelistedDirs = [
@@ -322,9 +323,7 @@ export const layer = Layer.effect(
           )
         }
 
-        const get = Effect.fnUntraced(function* (agent: string) {
-          return agents[agent]
-        })
+        const get = (agent: string) => Effect.succeed(agents[agent])
 
         const list = Effect.fnUntraced(function* () {
           const cfg = yield* config.get()

packages/opencode/src/command/index.ts

@@ -5,6 +5,7 @@ import type { InstanceContext } from "@/project/instance-context"
 import { SessionID, MessageID } from "@/session/schema"
 import { Effect, Layer, Context, Schema } from "effect"
 import { Config } from "@/config/config"
+import { Plugin } from "@/plugin"
 import { MCP } from "../mcp"
 import { Skill } from "../skill"
 import PROMPT_INITIALIZE from "./template/initialize.txt"
@@ -66,10 +67,12 @@ export const layer = Layer.effect(
   Service,
   Effect.gen(function* () {
     const config = yield* Config.Service
+    const plugin = yield* Plugin.Service
     const mcp = yield* MCP.Service
     const skill = yield* Skill.Service
 
     const init = Effect.fn("Command.state")(function* (ctx: InstanceContext) {
+      yield* plugin.init()
       const cfg = yield* config.get()
       const bridge = yield* EffectBridge.make()
       const commands: Record<string, Info> = {}
@@ -173,6 +176,7 @@ export const layer = Layer.effect(
 )
 
 export const defaultLayer = layer.pipe(
+  Layer.provide(Plugin.defaultLayer),
   Layer.provide(Config.defaultLayer),
   Layer.provide(MCP.defaultLayer),
   Layer.provide(Skill.defaultLayer),

packages/opencode/src/provider/transform.ts

@@ -119,6 +119,8 @@ function normalizeMessages(
           })
         }
         return msg
+      default:
+        return msg
     }
   })
 
@@ -334,7 +336,20 @@ function normalizeMessages(
     })
   }
 
-  return msgs
+  // Strip reasoning_content / reasoning_details from messages when the provider
+  // does not support interleaved reasoning. Otherwise providers like Cerebras
+  // reject the request with "property 'messages.N.assistant.reasoning_content' is unsupported".
+  return msgs.map((msg) => {
+    if (msg.role !== "assistant" || !msg.providerOptions?.openaiCompatible) return msg
+    const { reasoning_content: _rc, reasoning_details: _rd, ...rest } = msg.providerOptions.openaiCompatible
+    return {
+      ...msg,
+      providerOptions: {
+        ...msg.providerOptions,
+        openaiCompatible: rest,
+      },
+    }
+  })
 }
 
 function applyCaching(msgs: ModelMessage[], model: Provider.Model): ModelMessage[] {

packages/opencode/src/session/llm.ts

@@ -327,6 +327,17 @@ const live: Layer.Layer<
           })
         : undefined
 
+      l.info("params", {
+        temperature: params.temperature,
+        topP: params.topP,
+        topK: params.topK,
+        maxOutputTokens: params.maxOutputTokens,
+        variant: input.user.model.variant,
+        providerOptions: ProviderTransform.providerOptions(input.model, params.options),
+        limit: input.model.limit,
+        capabilities: input.model.capabilities,
+      })
+
       const opencodeProjectID = input.model.providerID.startsWith("opencode")
         ? (yield* InstanceState.context).project.id
         : undefined