@dreb/coding-agent 2.63.0 → 2.64.1

package/docs/rpc.md

@@ -18,6 +18,12 @@ await client.start();
 
 This enables per-user filesystem isolation by plain Unix DAC: give each authenticated user a dedicated UID and a working directory owned by that UID at mode `0700`. If the parent lacks the required capability (or the platform doesn't support `uid`/`gid`, e.g. Windows), the spawn fails and `start()` rejects rather than silently running as the parent user.
 
+### Process exit and pipe errors
+
+`RpcClient.onExit(listener)` receives an `RpcExitInfo` when the child dies: `{ code, signal }` from a process `exit`, or `{ error }` from a spawn/runtime `error`. Process exits also carry `stderrTail` — the last **2000 characters** of the child's captured stderr — so a host that only watches for exit codes can still surface the child's own diagnostic (for example the stdout backpressure guard's abort message) instead of an opaque exit code. In-flight requests are rejected with the exit reason.
+
+Child stdio pipe failures (e.g. an `EPIPE` when writing a large prompt to a dying child) are handled the same way: each pipe's `error` event fails in-flight requests with the pipe error and the captured stderr tail instead of crashing the host process.
+
 
 ## Starting RPC Mode
 
@@ -1325,7 +1331,7 @@ Note: with `summarize: true` the command is LLM-bound and can take a while. `Rpc
 
 Persistent settings, backed by the settings file (see [settings.md](settings.md)). They are normally distinct from live session state, with global-only control/security-policy exceptions:
 
-- **Persistent defaults** (`get_settings` / `set_settings`): provider/model, thinking level, queue modes, compaction/retry/image/skill/thinking-display/transport toggles, `maxConcurrentSubagents`, `enabledModels`, `tabTitle`, and per-agent model fallback lists seed fresh runtimes. Writing these ordinary defaults does **not** change a running session.
+- **Persistent defaults** (`get_settings` / `set_settings`): provider/model, thinking level, queue modes, compaction/retry/image/skill/thinking-display/transport toggles, automatic-compaction continuation, `maxConcurrentSubagents`, `enabledModels`, `tabTitle`, and per-agent model fallback lists seed fresh runtimes. Writing these ordinary defaults does **not** change a running session.
 - **Global nested-context trust policy** (`autoLoadNestedContext`, `trustedContextFolders`, `effectiveTrustedContextRoots`, and the trust commands below): this is read from `~/.dreb/agent/settings.json` only, never project settings. Active main/subagent processes observe it for **future lazy nested/out-of-cwd loads**; it cannot remove content already injected into a conversation. It does not govern the separate initial upward context scan from the launch cwd.
 - **Global Dispatch Arbiter policy** (`subagentArbiter`): the complete object is read/written globally and project settings cannot shadow it. Enabled runtimes consume it before future subagent spawns; it does not rewrite already-started children.
 - **Runtime state** (`get_state` / `set_model` / `set_thinking_level` / `set_steering_mode` / `set_follow_up_mode` / `set_auto_compaction` / `set_auto_retry`): the state of the live session. Note that the runtime setters also persist their values as new defaults as a side effect.
@@ -1353,6 +1359,7 @@ Response:
     "steeringMode": "one-at-a-time",
     "followUpMode": "one-at-a-time",
     "compactionEnabled": true,
+    "continueAfterAutoCompaction": false,
     "retryEnabled": true,
     "maxConcurrentSubagents": 4,
     "imageAutoResize": true,
@@ -1400,6 +1407,8 @@ Response:
 
 `tabTitle` is the effective merged object and is absent when unconfigured. Its optional `model` is one exact `provider/model`; when absent, title generation preserves the Explore-agent resolution route documented in [settings.md](settings.md#tab-title).
 
+`continueAfterAutoCompaction` defaults to `false`. When true, every successful automatic threshold or overflow compaction starts another model turn even without queued messages. It does not make failed, cancelled, or manual compaction continue.
+
 #### set_settings
 
 Update persistent default settings. Takes a partial payload — only the supplied keys change. The whole payload is validated before anything is applied: on any invalid field, nothing changes and the response is an explicit error. Writes target the global settings file (same scope as every runtime setter).
@@ -1408,6 +1417,12 @@ Update persistent default settings. Takes a partial payload — only the supplie
 {"type": "set_settings", "settings": {"defaultThinkingLevel": "low", "retryEnabled": false}}
 ```
 
+Enable uninterrupted automatic-compaction continuation for unattended work. This can keep producing model turns and cost indefinitely; it never changes manual `compact` behavior:
+
+```json
+{"type": "set_settings", "settings": {"continueAfterAutoCompaction": true}}
+```
+
 Set the maximum concurrency for newly started parent sessions. Zero removes the `subagent` tool and adds explicit parent-model guidance; positive values limit running children without changing the separate eight-item parallel-call input bound:
 
 ```json
@@ -1488,6 +1503,7 @@ Response is the full settings snapshot after the write (same shape as `get_setti
     "steeringMode": "one-at-a-time",
     "followUpMode": "one-at-a-time",
     "compactionEnabled": true,
+    "continueAfterAutoCompaction": false,
     "retryEnabled": false,
     "maxConcurrentSubagents": 1,
     "imageAutoResize": true,
@@ -1514,6 +1530,7 @@ Project-shadow warning example (the global write still lands, but the returned m
     "steeringMode": "one-at-a-time",
     "followUpMode": "one-at-a-time",
     "compactionEnabled": true,
+    "continueAfterAutoCompaction": false,
     "retryEnabled": true,
     "maxConcurrentSubagents": 4,
     "imageAutoResize": true,
@@ -1543,6 +1560,7 @@ Valid keys and values:
 | `steeringMode` | `"all"`, `"one-at-a-time"` |
 | `followUpMode` | `"all"`, `"one-at-a-time"` |
 | `compactionEnabled` | boolean |
+| `continueAfterAutoCompaction` | boolean; default `false`. Starts another model turn after every successful automatic compaction and never affects manual compaction. |
 | `retryEnabled` | boolean |
 | `maxConcurrentSubagents` | Non-negative safe integer; default `4`. Captured by new parent sessions; `0` removes the subagent tool and adds explicit self-execution guidance. |
 | `imageAutoResize` | boolean |
@@ -1859,6 +1877,14 @@ When the RPC server is launched with `--ui dashboard`, `message_update` events a
 
 The projection applies recursively to `message_update` events nested inside `background_agent_event` payloads. It does **not** apply to command responses — `get_dashboard_snapshot` still returns complete messages — and `message_end` always carries the full final message as the authoritative transcript record. RPC servers launched without `--ui dashboard` emit the full unprojected protocol shown above.
 
+The same dashboard-mode projection also dedupes images across the wire. Inline `image` blocks (PNG, JPEG, GIF, WebP) are content-identified by `sha256(mimeType + 0x00 + decodedBytes)`; the first occurrence of each unique image is sent inline, and every later occurrence anywhere in the event stream is replaced with
+
+```json
+{"type": "image_reference", "id": "<64 hex chars>", "mimeType": "image/png", "size": 12345}
+```
+
+The child process emits each unique image's bytes at most once per process lifetime, which keeps multi-image turns from filling the stdout pipe while the dashboard is busy decoding. The child only dedupes blocks the dashboard's strict decode accepts — allowlisted MIME type, canonical base64, matching byte signature; everything else is left inline at every occurrence (the dashboard rejects it, as before, so it never becomes an unresolvable reference). Command responses are untouched — `get_messages` and `get_dashboard_snapshot` always carry full base64 payloads — and the dedupe state is per-process: a restarted child re-sends. See [Transcript images](dashboard.md#transcript-images) for how the dashboard resolves these references.
+
 ### tool_execution_start / tool_execution_update / tool_execution_end
 
 Emitted when a tool begins, streams progress, and completes execution.
@@ -1928,7 +1954,7 @@ The `reason` field is `"threshold"` (context getting large) or `"overflow"` (con
 }
 ```
 
-If `reason` was `"overflow"` and compaction succeeds, `willRetry` is `true` and the agent will automatically retry the prompt.
+If `reason` was `"overflow"` and compaction succeeds, `willRetry` is `true` and the agent will automatically retry the prompt. Independently, persistent `continueAfterAutoCompaction: true` starts another model turn after any successful automatic compaction, including a threshold compaction with `willRetry: false` and no queued message. Manual `compact` is outside this event path and does not continue because of that setting.
 
 If compaction was aborted, `result` is `null` and `aborted` is `true`.
 

package/docs/settings.md

@@ -215,6 +215,7 @@ For either permitted lazy-load path, each context file is realpath-deduplicated
 | Setting | Type | Default | Description |
 |---------|------|---------|-------------|
 | `compaction.enabled` | boolean | `true` | Enable auto-compaction |
+| `compaction.continueAfterAutoCompaction` | boolean | `false` | Start another model turn after every successful automatic compaction; manual `/compact` never continues because of this setting |
 | `compaction.reserveTokens` | number | `16384` | Tokens reserved for LLM response |
 | `compaction.keepRecentTokens` | number | `20000` | Recent tokens to keep (not summarized) |
 
@@ -222,12 +223,15 @@ For either permitted lazy-load path, each context file is realpath-deduplicated
 {
   "compaction": {
     "enabled": true,
+    "continueAfterAutoCompaction": false,
     "reserveTokens": 16384,
     "keepRecentTokens": 20000
   }
 }
 ```
 
+The continuation option is intended for unattended long-running work and can produce model turns and cost without further user input. It applies only after successful automatic threshold or overflow compaction. Failed or cancelled automatic compaction and manual `/compact` do not continue because of it. The option is available in terminal `/settings` and dashboard Settings.
+
 ### Branch Summary
 
 | Setting | Type | Default | Description |
@@ -464,6 +468,7 @@ See [packages.md](packages.md) for package management details.
   "theme": "dark",
   "compaction": {
     "enabled": true,
+    "continueAfterAutoCompaction": false,
     "reserveTokens": 16384,
     "keepRecentTokens": 20000
   },

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@dreb/coding-agent",
-	"version": "2.63.0",
+	"version": "2.64.1",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"type": "module",
 	"drebConfig": {