@dreb/coding-agent 2.25.3 → 2.27.2

package/CHANGELOG.md

@@ -4,6 +4,8 @@
 
 ### Added
 
+- Thinking summaries are visible again on Opus 4.7+ (and any adaptive-thinking Claude model). Anthropic flipped the API default for `thinking.display` to `"omitted"`, so these models returned empty thinking blocks — dreb now sends `"summarized"` by default. Added a model-keyed `modelSettings` setting with a `thinkingDisplay` override (`"summarized" | "omitted"`), exposed in `/settings` → **Show thinking summaries** (shown only for adaptive models). Because the setting is keyed by model ID and resolved at the shared session-creation chokepoint, it is honored uniformly across the TUI, headless, JSON, RPC, and Telegram clients — and inherited automatically by subagents using the same model. ([#250](https://github.com/aebrer/dreb/issues/250))
+
 - Secret scrubbing and sensitive file access guards — two layers of defense against accidental credential leaks through the tool pipeline. Output scrubbing detects and redacts known secret patterns (AWS, GitHub, OpenAI, Anthropic, Slack, Stripe, PEM/SSH keys, URL credentials) in tool output before it enters the LLM conversation. Sensitive file guard blocks read access to credential files (`~/.ssh/id_*`, `~/.aws/credentials`, `~/.dreb/secrets/`, etc.) via both the `read` tool and bash commands. Both layers configurable via `sensitiveFilePaths` and `secretOutputPatterns` settings. ([#171](https://github.com/aebrer/dreb/issues/171))
 
 - Expanded forbidden-commands guard with destructive operation patterns: `rm -rf /` and variants, `dd` to block devices, `mkfs`, fork bomb `:(){ :|:& };:`, and block device redirects (`> /dev/sda`). Guard now also checks quoted content (catches `echo "rm -rf /" | bash`) and inspects script files before execution (`bash script.sh`). ([#170](https://github.com/aebrer/dreb/issues/170))
@@ -40,6 +42,8 @@
 
 ### Fixed
 
+- Fixed ESC doing nothing during `stream_retry` / `length_retry` backoff. The old handlers stopped the working spinner (`loadingAnimation`) and swapped in a dedicated retry spinner that had no ESC wiring, so the default ESC handler's `if (this.loadingAnimation)` guard fell through to a no-op. The fix removes the retry spinner entirely, keeps the working spinner running throughout the retry, and surfaces retry feedback as a persistent warning in the chat scrollback instead. ESC now aborts via the pre-existing spinner-abort path — identical to cancelling normal generation. ([#262](https://github.com/aebrer/dreb/issues/262))
+
 - Fixed `/buddy off` not persisting across sessions — the buddy reappeared in the TUI on restart. The `hidden` flag was persisted correctly, but the startup auto-mount ignored it. The startup mount is now gated on the persisted `hidden` flag (extracted into `mountExistingBuddyIfVisible()`); a hidden buddy stays loaded but unmounted until you run `/buddy` to bring it back. ([#243](https://github.com/aebrer/dreb/issues/243))
 
 - Fixed subagents silently returning empty results when a child's final response was truncated at the model's token limit. The subagent spawn handler now inspects the last assistant message's `stopReason`: a clean exit with no output now surfaces a descriptive error (truncation, loud truncation failure, or "completed with no output") instead of a silent empty string, and a clean exit with partial output now flags the truncation via `errorMessage` while preserving the partial text — covering both raw `length` truncation and the loud `error` stopReason produced when the core agent loop exhausts its length retries. The parent renders these errors even on a clean (exit code 0) exit. ([#240](https://github.com/aebrer/dreb/issues/240))

package/README.md

@@ -43,6 +43,22 @@ What you get in exchange: a skill system, an extension API, custom agent definit
 
 ## Quick Start
 
+> **Node.js 22 LTS is required.** dreb relies on SSE streaming behavior that is stable in Node 22 LTS. Node 24 and Node 26 are known to break provider streaming due to changes in ReadableStream buffering, which causes every provider to fail with **"request ended without sending any chunks"**. If you see that error, switch to Node 22 LTS.
+
+### Building from source (recommended)
+
+```bash
+git clone https://github.com/aebrer/dreb.git
+cd dreb
+npm install
+npm run build
+npm link -w packages/coding-agent
+```
+
+---
+
+### Installing from npm
+
 ```bash
 npm install -g @dreb/coding-agent
 ```
@@ -71,17 +87,9 @@ Then just talk to dreb. All 11 built-in tools are enabled by default: `read`, `w
 
 **Bun users:** Bun's lockfile can cache stale `@dreb/*` versions after upgrades, causing missing-export errors. Fix with `bun pm cache rm && bunx --force dreb`.
 
----
+### Troubleshooting
 
-### Building from source
-
-```bash
-git clone https://github.com/aebrer/dreb.git
-cd dreb
-npm install
-npm run build
-npm link -w packages/coding-agent
-```
+- **"request ended without sending any chunks" on every provider** — Your Node version is likely too new. Switch to **Node.js 22 LTS**. Node 26 in particular changed ReadableStream buffering in a way that breaks the Anthropic and OpenAI SDK stream parsers dreb uses.
 
 ---
 
@@ -157,7 +165,7 @@ Type `/` in the editor to trigger commands. [Extensions](#extensions) can regist
 | `/login`, `/logout` | OAuth authentication |
 | `/model` | Switch models |
 | `/scoped-models` | Enable/disable models for cycling |
-| `/settings` | Thinking level, theme, message delivery, transport |
+| `/settings` | Thinking level, thinking summaries (adaptive Claude models), theme, message delivery, transport |
 | `/resume` | Pick from previous sessions |
 | `/new` | Start a new session |
 | `/name <name>` | Set session display name |
@@ -361,6 +369,8 @@ The `subagent` tool delegates tasks to independent child agent processes. Each s
 
 **Per-agent model overrides:** The model used by each agent type can be overridden via the `agentModels.models` setting (a map of agent name → ordered fallback list) without copying or editing the agent definition `.md` files. Configure it in `settings.json` or via `/settings` → **Agent Models**. Resolution order: per-invocation `model` override → `agentModels` setting → agent definition `model` → parent session model. See [docs/agent-models.md](docs/agent-models.md).
 
+**Model identity in system prompt:** The parent session's running model is exposed in its own system prompt as `You are running on: provider/id`. This lets the model make self-aware routing decisions (e.g. delegate vision tasks to a multimodal subagent, or use a differently-architected model as a critic). The line updates automatically on mid-session model switches.
+
 **Session metadata:** Each child process records its agent type in the session JSONL header (`agentType` field), providing an audit trail of which agent definition executed the work.
 
 ---

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

@@ -467,6 +467,13 @@ export declare class AgentSession {
      * @throws Error if no API key available for the model
      */
     setModel(model: Model<any>): Promise<void>;
+    /**
+     * Refresh the live agent's thinkingDisplay for a (possibly) new model. The default
+     * depends on the model (adaptive-thinking models default to "summarized"), so this
+     * must run on every model switch. Keyed by model id from shared settings, so it
+     * resolves identically to how createAgentSession seeds it at startup.
+     */
+    private _refreshThinkingDisplay;
     /**
      * Cycle to next/previous model.
      * Uses scoped models (from --models flag) if available, otherwise all available models.