@dreb/coding-agent 2.59.1 → 2.60.0

package/README.md

@@ -81,7 +81,7 @@ Or use a custom provider (corporate proxy, Bedrock, etc.) — see [Custom provid
 
 Then just talk to dreb. All 13 standard built-in tools are enabled by default (unless `backgroundAgents.maxConcurrentSubagents` is `0`, which removes `subagent` from new parent sessions): `read`, `write`, `edit`, `bash`, `grep`, `find`, `ls`, `web_search`, `web_fetch`, `subagent`, `wait`, `watch_github_ci`, and `ask_user`. Use `--tools` to restrict to a subset (e.g., `--tools read,grep,find,ls` for read-only). Three additional tools — `search`, `skill`, and `tasks_update` — are always active regardless of `--tools`. `suggest_next` is active by default but excluded when `--tools` is specified. The model uses these to fulfill your requests. Add capabilities via [skills](#skills), [prompt templates](#prompt-templates), [extensions](#extensions), or [packages](#packages).
 
-**Also available:** [`@dreb/telegram`](https://www.npmjs.com/package/@dreb/telegram) — run dreb as a Telegram bot with live tool status and visible results for user-facing tools (`npm install -g @dreb/telegram`). [`@dreb/dashboard`](https://www.npmjs.com/package/@dreb/dashboard) — run `dreb dashboard` for a browser UI with fleet overview, full chat steering, unified dismissible session banners, explicit navigation that preserves viewed closed transcripts as read-only snapshots, a TUI-parity rolling TPS indicator with long-term delta, generic fail-closed built-in slash-command discovery and execution, inline provider/API failures with partial output preserved, sanitized raster tool images plus sent user uploads retained as bounded transcript previews by default, a bounded all-agent subagent panel with drill-in, host file browser, dreb memory editor with exact-revision saves and automatic index cleanup on delete, curated appearance themes (per-browser light/dark), and resilient Tailscale/rotating-code pairing with a configurable 180-day default and advance expiry warnings (`npm install -g @dreb/dashboard`; see [docs/dashboard.md](docs/dashboard.md)). Tool images cross browser-facing transport as content-addressed references; browser-local Settings offers placeholders, bounded previews, or informed-opt-in originals, with size disclosure and confirmation above 1 MiB. Full-resolution HTML export remains self-contained. The Memories screen is dreb-only (`~/.dreb/memory` and populated active/on-disk-session project `.dreb/memory`; empty projects are omitted), shows complete indexes with a >200-line warning, opens local index links within the selected scope, replaces stale editor content with visible loading feedback, surfaces malformed entry frontmatter for repair, preserves drafts on conflicts, and does not create/rename entries or expose Claude paths. Compact SSE snapshots update live fleet cards without repeatedly fetching the cross-project inventory, and session drill-in hydrates state, messages, and background agents through one ordered snapshot request. Terminal provider failures show their reason on fleet cards, while transient failures clear terminal state when automatic retry begins and remain recorded inline on the failed attempt. Its top bar and persistent session header indicators report connecting, connected, retrying, resyncing, disconnected, or auth failed; bounded SSE replay plus an explicit snapshot barrier restores session state, tasks, and image references after a reload, restart, gap, backpressure disconnect, or stalled stream, while authenticated image routes recover bytes separately from authoritative transcripts.
+**Also available:** [`@dreb/telegram`](https://www.npmjs.com/package/@dreb/telegram) — run dreb as a Telegram bot with live tool status and visible results for user-facing tools (`npm install -g @dreb/telegram`). [`@dreb/dashboard`](https://www.npmjs.com/package/@dreb/dashboard) — run `dreb dashboard` for a browser UI with fleet overview, full chat steering, unified dismissible session banners, explicit navigation that preserves viewed closed transcripts as read-only snapshots, a TUI-parity rolling TPS indicator with long-term delta, generic fail-closed built-in slash-command discovery and execution, inline provider/API failures with partial output preserved, sanitized raster tool images plus sent user uploads retained as bounded transcript previews by default, a bounded all-agent subagent panel with drill-in, host file browser, dreb memory editor with exact-revision saves and automatic index cleanup on delete, curated appearance themes (per-browser light/dark), and resilient Tailscale/rotating-code pairing with a configurable 180-day default and advance expiry warnings (`npm install -g @dreb/dashboard`; see [docs/dashboard.md](docs/dashboard.md)). Tool images cross browser-facing transport as content-addressed references; browser-local Settings offers placeholders, bounded previews, or informed-opt-in originals, with size disclosure and confirmation above 1 MiB. Full-resolution HTML export remains self-contained. The Memories screen is dreb-only (`~/.dreb/memory` and populated active/on-disk-session project `.dreb/memory`; empty projects are omitted), shows complete indexes with a >200-line warning, opens local index links within the selected scope, replaces stale editor content with visible loading feedback, surfaces malformed entry frontmatter for repair, preserves drafts on conflicts, and does not create/rename entries or expose Claude paths. Compact SSE snapshots update live fleet cards without repeatedly fetching the cross-project inventory, and session drill-in hydrates state, messages, and background agents through one ordered snapshot request. While drill-in remains mounted, authoritative detail refreshes keep its header current, confirmed model/thinking changes are reconciled without stale snapshot rollback, and post-compaction context uses a conservative estimate until fresh provider usage arrives. Terminal provider failures show their reason on fleet cards, while transient failures clear terminal state when automatic retry begins and remain recorded inline on the failed attempt. Its top bar and persistent session header indicators report connecting, connected, retrying, resyncing, disconnected, or auth failed; bounded SSE replay plus an explicit snapshot barrier restores session state, tasks, and image references after a reload, restart, gap, backpressure disconnect, or stalled stream, while authenticated image routes recover bytes separately from authoritative transcripts.
 
 **Platform notes:** [Windows](docs/windows.md) | [Termux (Android)](docs/termux.md) | [tmux](docs/tmux.md) | [Terminal setup](docs/terminal-setup.md) | [Shell aliases](docs/shell-aliases.md)
 
@@ -319,6 +319,8 @@ A trusted root permits lazy loading for that existing directory and all of its d
 
 Replace the default system prompt with `.dreb/SYSTEM.md` (project) or `~/.dreb/agent/SYSTEM.md` (global). Append without replacing via `APPEND_SYSTEM.md`.
 
+For persistent instructions that apply only to one exact provider/model pair, use `modelSettings` in `settings.json` with a canonical key such as `openai-codex/gpt-5.6-sol`. Set `systemPrompt` to replace dreb's built-in prompt or `appendSystemPrompt` to append instructions. The same mechanism works for local/custom models registered in `models.json`, and the prompt updates immediately when the model changes. Explicit session replacements (`--system-prompt`, `SYSTEM.md`, or SDK hooks) take precedence. See [Model settings](docs/settings.md#modelsettings).
+
 ---
 
 ## Memory

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

@@ -23,7 +23,8 @@ import type { ModelRegistry } from "./model-registry.js";
 import { PerformanceTracker } from "./performance-tracker.js";
 import { type PromptTemplate } from "./prompt-templates.js";
 import type { ResourceLoader } from "./resource-loader.js";
-import type { BranchSummaryEntry, SessionManager } from "./session-manager.js";
+import type { BranchSummaryEntry } from "./session-manager.js";
+import { SessionManager } from "./session-manager.js";
 import { type SettingsManager } from "./settings-manager.js";
 import type { BashOperations } from "./tools/bash.js";
 import { type SessionTask, type SubagentArbitrationEvent, type SubagentResult, type SubagentStepMetadata } from "./tools/index.js";
@@ -562,6 +563,8 @@ export declare class AgentSession {
      * resolves identically to how createAgentSession seeds it at startup.
      */
     private _refreshThinkingDisplay;
+    /** Reject malformed target-model prompt settings before a model switch mutates session state. */
+    private _validateModelPromptSettings;
     /**
      * Cycle to next/previous model.
      * Uses scoped models (from --models flag) if available, otherwise all available models.