@leanandmean/coding-agent 0.74.1-scramjet.6 → 0.74.1-scramjet.8

package/CHANGELOG.md

@@ -4,11 +4,13 @@
 
 ### Added
 
+- Custom `models.json` schema support for Anthropic Messages `supportsTemperature` and `forceAdaptiveThinking` compat fields used by the Opus 4.8 / Fable 5 / Sonnet 5 model support update, with effective-API validation that rejects compat fields outside the target API contract.
 - `beforeToolBatch` hook on `AgentLoopConfig` / `Agent` — called after assistant `message_end` but before tool-call extraction, allowing async event handlers (e.g., queued `message_end` mutations) to settle before the execution pipeline reads tool calls ([#196](https://github.com/LeanAndMean/scramjet/issues/196))
 - `AgentSession._drainAgentEventQueue()` helper — awaits the queued extension event promise; wired into both `beforeToolBatch` (pre-extraction drain) and `beforeToolCall` (per-tool drain)
 
 ### Changed
 
+- Default Anthropic model resolution now selects `claude-opus-4-8`; default Amazon Bedrock model resolution now selects `us.anthropic.claude-opus-4-8`. Unknown custom CLI model IDs no longer inherit model-specific `compat` or `thinkingLevelMap` metadata from those defaults.
 - **Documentation rebrand**: Rebranded all shipped documentation, examples, and system prompt from Pi identity to Scramjet identity. Product name references, binary name (`pi` → `scramjet`), env var names (`PI_*` → `SCRAMJET_*`), and install instructions updated throughout. API variable names (`pi.on()`, `pi.registerTool()`) and `.pi/` config directory paths are preserved. ([#199](https://github.com/LeanAndMean/scramjet/issues/199))
 
 - Merged upstream v0.74.1: Together AI provider, image generation APIs, Node 26 fetch fixes, session affinity headers, provider compat improvements ([#14](https://github.com/LeanAndMean/pi/issues/14))

package/dist/core/agent-session.d.ts

@@ -16,7 +16,7 @@ import type { Agent, AgentEvent, AgentMessage, AgentState, AgentTool, ThinkingLe
 import type { ImageContent, Model, TextContent } from "@leanandmean/ai";
 import { type BashResult } from "./bash-executor.js";
 import { type CompactionResult } from "./compaction/index.js";
-import { type ContextUsage, type DeliverAs, type DispatchUserInputOptions, type ExtensionCommandContextActions, type ExtensionErrorListener, ExtensionRunner, type ExtensionUIContext, type InputSource, type ReplacedSessionContext, type SendMessageDeliverAs, type SessionStartEvent, type ShutdownHandler, type ToolDefinition, type ToolInfo } from "./extensions/index.js";
+import { type ContextUsage, type DeliverAs, type DispatchUserInputOptions, type ExtensionCommandContextActions, type ExtensionErrorListener, ExtensionRunner, type ExtensionUIContext, type InputSource, type InvokeHarnessToolOptions, type ReplacedSessionContext, type SendMessageDeliverAs, type SessionStartEvent, type ShutdownHandler, type ToolDefinition, type ToolInfo } from "./extensions/index.js";
 import type { CustomMessage } from "./messages.js";
 import type { ModelRegistry } from "./model-registry.js";
 import { type PromptTemplate } from "./prompt-templates.js";
@@ -285,6 +285,19 @@ export declare class AgentSession {
      * Changes take effect on the next agent turn.
      */
     setActiveToolsByName(toolNames: string[]): void;
+    /**
+     * Execute a registered tool as a harness-originated call (not requested by the model).
+     *
+     * Resolves the wrapped tool from the full registry — including `"harness-only"` tools that are
+     * never provider-visible — so its `execute` still receives the extension context and its
+     * `tool_call`/`tool_result` hooks still fire. Delegates to {@link Agent.runHarnessTool}, which
+     * runs the call through the real prepare/execute/finalize pipeline: identical live
+     * `tool_execution_*`/message events, persisted session entries, and extension hooks, but no
+     * run/turn framing. The idle-vs-mid-run branch lives in `runHarnessTool` (keyed on the agent's
+     * active run, which tracks `isStreaming`): idle calls execute immediately, mid-run calls queue
+     * and drain before the next intra-run LLM call.
+     */
+    invokeHarnessTool(name: string, args: unknown, options?: InvokeHarnessToolOptions): Promise<void>;
     /** Whether compaction or branch summarization is currently running */
     get isCompacting(): boolean;
     /** All messages including custom types like BashExecutionMessage */