Hmbown/CodeWhale
61.3
Adequate · 14 November 2026
1528.2k
lines of production code
Rust
primary language
6
measurements over time
What this system is
Features
Add Nix package for CodeWhale
Introduces a new Nix derivation (nix/package.nix) that builds the CodeWhale Rust application. The package configures build inputs for OpenSSL and D-Bus, sets up native check inputs for testing, and implements a custom check phase to handle sandboxed test execution, including setting up temporary HOME directories, configuring SSL certificates, and patching runtime RPATHs for Linux-specific dependencies like libdbus.
nix · high confidence
Added Tencent Lighthouse deployment templates and systemd services
This change introduces a complete deployment configuration set for Tencent Lighthouse, including CNB pipeline templates (README, example configs, and deploy buttons), systemd service units for the CodeWhale runtime and Feishu/Telegram bridges, and example environment files for runtime, Feishu, and Telegram bridges. Users can now deploy CodeWhale to Tencent Lighthouse using these provided templates, which handle service management, environment variable configuration, and CI/CD integration via CNB.
deploy · high confidence
Added performance probe for session serialization and I/O
A new example tool, \zz\_perf\_probe.rs\, has been added to the TUI examples to benchmark the performance of session data handling. This utility measures the time required for deserializing session files into typed structures, serializing them back to JSON (both with and without message data), cloning session objects, converting journals to messages, and performing atomic file writes with fsync. This allows developers to profile the efficiency of the underlying \codewhale\_models\ and \codewhale\_runtime\ session tree operations.
crates/tui/examples · high confidence
Android port of the Codewhale pet simulation
The pet simulation is now available on Android, bringing the deterministic audiovisual world to the platform. This release includes the core simulation engine (PetSim), a native audio output layer, a bounded and atomic habitat store for recordings, and a shared-client mode that connects to a local companion via an adb reverse tunnel. Users can now run the pet in Wild, Demo, Recording, Live, and Shared modes, with support for saving, exporting, and archiving telemetry data.
android · high confidence
Codewhale CLI rebrand and account-centric command surface
The CLI binary is now named \codewhale\ (with \codew\ kept as a compatibility alias) and introduces a dedicated \codewhale cloud\ command group for managing the user's managed account. This new surface handles account sign-in via device-code flow, remote credential vaulting, and project/agent management, separating account identity from local provider configuration. The update also adds a \codewhale dispatch\ command for remote agent task execution, a \codewhale metrics\ command for reading usage receipts, and a \codewhale config\ bundle system for portable, secret-free configuration import/export.
crates/cli/src · high confidence
DeepSeek Harness integration with Codewhale branding and ocean scene
The TUI now includes a new integration module for DeepSeek Harness (DSH) that installs a dedicated 'codewhale' profile and applies a custom Codewhale skin via DSH's \overrideTokens\ mechanism. This integration mounts a React-based brand lockup (featuring a whale silhouette SVG and 'WHALE BROTHERS' text) into DSH's overlay slot and injects an ambient 'ocean scene' canvas with animated whales and fish behind the DSH web UI. The module also handles DSH detection, version compatibility checking, and secure identity mapping for routes, ensuring the user's existing DSH profiles and credentials remain untouched.
crates/tui/src/integrations/dsh · high confidence
Introduce AppMode and provider auth source configuration
The config crate now defines the TUI's operating modes (Plan, Act, Operate) with a dedicated cycle and legacy alias handling, and introduces a new \auth\_source\ schema allowing providers to fetch credentials via external commands or secret IDs rather than static keys.
crates/config/src · high confidence
Introduce Codewhale npm launcher with platform binary downloads and verification
The \npm/codewhale\ package is introduced as a Node.js wrapper that downloads, verifies, and launches the native Codewhale CLI/TUI binary. It supports Linux (x64, arm64), macOS (x64, arm64), Windows (x64, arm64), and Android arm64 (Termux), with HarmonyOS treated as Linux. The installer probes GitHub Releases and a CNB (China-friendly) mirror for Linux x64, locking to the first source that validates against the release SHA-256 manifest. It includes a glibc preflight check for Linux binaries, configurable download timeouts and retries, and environment variables to override the release source, version, and mirror behavior. The package also ships tests for artifact detection, compiled-host catalog validation, and install/verification flows.
npm · high confidence
Introduce Codewhale pet: a deterministic audiovisual activity visualization
Adds a new 'pet' module that visualizes agent activity as a persistent, 980-particle 'whale' form. This feature includes a shared local companion process that manages the world state, deterministic audio (stereo 48 kHz PCM), and recording (JSONL segments). It provides multiple view clients: a Ratatui terminal habitat, a browser-based viewer, a native macOS window, and new Android (Compose) and iOS (Swift) mobile habitats. The pet visualizes telemetry categories like editing, testing, and browsing, with versioned particle behaviors (v1 and v2) and configurable appearance presets. It supports live recording from the Runtime API, checkpoint persistence, and cross-platform conformance testing.
pet · high confidence
Introduce Feishu bridge integration
Adds a new Feishu (Lark) bridge integration that connects the platform to Feishu chats via the official Lark SDK. The bridge supports direct messages and configurable group chats, allowing users to interact with the runtime through text commands (such as /model, /new\thread, /resume) and reply within existing threads/topics rather than creating new standalone conversations. It handles authentication, message parsing, and state persistence, with configuration driven by environment variables (including fallbacks from DEEPSEEK\\* to CODEWHALE\\ prefixes) and strict allowlist controls for chat access.
integrations/feishu-bridge, integrations/telegram-bridge · high confidence
Introduce Lane runtime with tmux and inline backends
The \crates/lane\ crate now provides the core runtime for executing workflow instances (Lanes). It introduces a typed control-plane contract for slash commands and CLI entrypoints, a durable registry for lane state under \$CODEWHALE\_HOME/lanes/\, and two execution backends: \tmux\ for detached sessions and \inline\ for direct process execution. The system manages isolated git worktrees for each lane, including automatic cleanup based on TTL, and ensures secure handling of secrets and paths by refusing linked worktree directories and redacting sensitive data in receipts.
crates/lane · high confidence
Introduce StateStore for historical thread recovery and canonical runtime receipts
The new \crates/state\ crate provides a \StateStore\ that manages historical thread archives and canonical runtime links via SQLite. It enables restoring immutable legacy conversation graphs in a single transaction and reading full thread history with branch and leaf information. Additionally, it publishes and validates canonical runtime receipts, ensuring that thread ownership and execution scope are correctly linked and preventing silent data loss through serialized session-index writes and lock compaction.
crates/state/src · high confidence
Introduce cloud facts client with signed envelope verification and background caching
Added a new \cloud-facts\ crate that fetches signed cloud facts from a remote endpoint, verifies the Ed25519 envelope against pinned keys, caches the verified payload to disk, and installs it as a process-wide overlay. The feature is disabled by default and can be enabled via the \CODEWHALE\_CLOUD\_FACTS\ environment variable or configuration, with a hard kill switch via \CODEWHALE\_DISABLE\_CLOUD\_FACTS\. It includes a proof-of-concept example (\fetch\_live\) for dogfooding and comprehensive tests covering verification, caching, and overlay installation.
crates/cloud-facts · high confidence
Introduce community assistant automation and maintainer review panel
The website now includes a community assistant system that uses Cloudflare Cron Triggers to draft issue triage, PR reviews, stale nudges, and weekly digests via the DeepSeek API. These drafts are stored in Workers KV and require explicit maintainer approval before being posted to GitHub. A new bilingual maintainer panel at /admin allows reviewing, editing, posting, or discarding these drafts, secured by a MAINTAINER\_TOKEN. The site also adds a changelog page, a community page, and a locale-aware catch-all 404, alongside new environment configuration and documentation for the new automation. Also in this area: 19 commits (6 fixes) modifying web/app; 3 commits (1 fix) modifying web/lib; 1 commit (1 fix) fixing web/public; 4 commits (3 fixes) fixing web/public.
web · low confidence · unverified
Introduce exact Fleet schema, workflow elevation assessment, and lane-scoped gates
The workflow crate now supports a new 'exact' Fleet schema that freezes provider/model choices at save time, rejecting late-binding selectors to ensure reproducible runs. It adds a pure, UI-free elevation assessment module that analyzes workflow plans to determine if operator approval is required based on capabilities like writes, shell access, and token budget. Additionally, lane-scoped gate nodes and handoffs are introduced to manage role transitions, allowing workflows to block, retry, or escalate based on verification and review outcomes.
crates/workflow/src · high confidence
Introduce first-party telemetry ingest Worker with IP-stripped analytics and optional PostHog forwarding
The \telemetry-ingest\ Cloudflare Worker is now deployed at \https://telemetry.codewhale.net/v1/telemetry\ to accept telemetry batches and write them to Workers Analytics Engine. The ingest is designed to be IP-stripped: it reads only \content-type\ and \content-length\ headers, never accesses client IP or geo metadata, and enforces a closed schema (v3 with notice version 5, or legacy v2 with consent version 4) that rejects any unknown fields. The worker supports an optional PostHog processor that forwards consented batches to PostHog, but this is disabled by default and requires explicit operator configuration of secrets and egress verification. Additionally, a new CLI script \report-active-installs.mjs\ allows operators to query observed active installs and session trends from the Analytics Engine data. Also in this area: 1 commit (0 fixes) modifying telemetry-ingest.
telemetry-ingest · medium confidence · unverified
Introduce interactive chat sidebar and runtime API client
The VS Code extension now features an interactive chat sidebar that connects to the CodeWhale Engine Runtime API. This change introduces a new HTTP/SSE client (\api.ts\) to manage threads, turns, and streaming events, alongside a \ChatView\ that handles the user interface, context chips (selection, file, diagnostics), and inline approvals. It also includes a dependency-free Markdown renderer for safe transcript display and a new \RuntimeStatusView\ to monitor connection state and list restore points. Additionally, token resolution has been updated to prioritize VS Code SecretStorage over legacy settings for improved security.
extensions/vscode · high confidence
Introduce lifecycle outbox for JSONL logging and webhook delivery
The \crates/hooks\ module now provides a \LifecycleOutbox\ that records session, turn, and subagent lifecycle events to a local JSONL file and optionally fans them out to a webhook. Events are written as structured JSON lines with monotonic sequence numbers and bounded payload fields (headline ≤ 80 chars, detail ≤ 120 chars, preview ≤ 200 chars) to prevent unbounded memory growth. The outbox uses an internal writer task to batch appends and limits concurrent webhook posts to eight in-flight requests, ensuring that stalled webhooks do not block the audit log. Queue capacity is capped at 1,024 events; excess events are dropped with a warning rather than blocking the caller. File writes are serialized via a mutex to prevent interleaved JSON lines, and parent directories are created with restrictive permissions (0700/0600) on Unix. A \flush\ method allows graceful shutdown to wait for pending deliveries within a timeout.
crates/hooks · high confidence
Introduce portable command-contract facets and shared formatting utilities
The \crates/command-contract\ crate now defines the object-safe capability interfaces (facets) and handler transport envelope that decouple command logic from the TUI host, enabling staged migration of command groups into a portable \codewhale-commands\ crate. This change introduces a typed capability bitmask (e.g., \SESSION\, \MODEL\, \DEBUG\_DIAGNOSTICS\) to enforce least-authority access for command handlers, alongside shared, host-agnostic formatting utilities for elapsed time, session metrics, monetary costs, and tool output status. It also establishes the \CommandResult\ and \DebugAction\ types that standardize how commands report outcomes and request host actions.
crates/command-contract/src · high confidence
Introduce sandboxed JavaScript Workflow runtime with typed error handling and schema repair
This change adds the \workflow-js\ crate, providing a sandboxed QuickJS runtime for executing imperative workflow scripts. The runtime introduces typed error classification (e.g., \admission\, \budget\, \schema\) to replace fragile text-based error parsing, ensuring reliable fan-out behavior in \parallel()\ and \pipeline()\ calls. It also implements bounded \responseSchema\ repair, allowing the system to automatically retry subagent calls that return invalid JSON or fail schema validation. The entry includes the core driver seam for task dispatch, error definitions, schema validation logic, and a \FakeDriver\ for testing without real subagents.
crates/workflow-js/src · high confidence
Introduce shared settings-picker framework with unified navigation and filtering
The TUI now uses a new shared \settings\_picker\ module to manage settings selection, providing a consistent user experience across different configuration areas. This framework introduces stable tab-based filtering and full-text search that spans all tabs, allowing users to quickly locate options. It standardizes keyboard navigation (including vim-style j/k keys and digit shortcuts) and enforces a uniform preview/commit/cancel lifecycle, ensuring that changes are reviewed before being applied. Additionally, the layout engine now automatically adapts to terminal width, switching between a side-by-side list/detail view and a stacked list-only view on narrow screens, with specific options able to force the list-only mode.
_crates/tui/src/tui/settings\picker · high confidence
Introduce signed, versioned cloud facts overlay for model catalog and provider defaults
A new \cloud\_facts\ module in the config crate enables a signed, versioned overlay that can update the model catalog and provider defaults without requiring a new binary release. The system uses pinned Ed25519 trust anchors (with a production key pinned in \keys.rs\) to verify incoming envelopes, ensuring that only authenticated patches are applied. These verified facts are scoped to the current binary version and time validity before being applied to the catalog via \catalog\_patch.rs\, which supports upserting, deprecating, or hiding model rows while preserving higher-layer configuration. The overlay is process-wide, off-by-default, and can be disabled via the \CODEWHALE\_DISABLE\_CLOUD\_FACTS\ environment variable, providing a secure mechanism for CodeWhale to push model corrections, pricing updates, and release announcements.
_crates/config/src/cloud\facts · high confidence
Introduce the \`crates/protocol\` library with core serializable types and contracts
A new \crates/protocol\ library has been added to centralize the serializable protocol contracts shared across the TUI, CLI, and app-server. This crate defines the foundational types and schemas that enable consistent communication and state projection across different surfaces. Key additions include \agent\_mail.rs\ for durable, scoped agent-to-agent messaging with strict size and control-character boundaries; \agent\_run.rs\ for a dependency-neutral read model of agent runs (covering lifecycle states, budgets, and terminal outcomes); \fleet.rs\ for the Agent Fleet control-plane protocol, including run handles, task/worker specs, and runtime events; \event\_msg.rs\ for the \EventMsg\ API that mirrors the engine's event variants to ensure byte-identical event shapes for headless and TUI clients; \engine\_owner.rs\ for a safe, ordered projection of engine owner activity; \ids.rs\ for typed \ThreadId\ and \SessionId\ wrappers; and \journal.rs\ for an append-only session tree journal with branching support. These types form the backbone of the protocol layer, ensuring that data crossing boundaries is serializable, validated, and consistent.
crates/protocol/src · high confidence
Introduce the unified Work Surface (workbar) with multi-panel dock and configurable placement
The TUI now features a unified Work Surface (also called the workbar) that consolidates session activity into a single, configurable component. Users can place this surface at the Bottom (default), Top, Left, or Right, or hide it entirely. The surface includes a dock with nine switchable panels—Tasks, Agents, Background, Terminal, Files, Notepad, Context, Git, and Price—accessible via keyboard cycling (Ctrl+Tab) or tab clicks. It provides persistent, selectable rows for todos and sub-agents that remain visible after completion, along with a live terminal view and session metadata (context budget, git status, cost). The component handles its own focus, selection, and detail inspection, replacing the previous fragmented sidebar and top-bar implementations.
_crates/tui/src/tui/work\surface · high confidence
Introduces a declarative route resolution and authentication contract
The configuration layer now enforces a strict route resolution contract where executable routes are produced exclusively by a central \RouteResolver\ and returned as immutable \ReadyRouteCandidate\ objects. This change introduces a new declarative authentication system (\AuthKind\, \AuthMethod\) that defines credential acquisition methods (API key, OAuth, keyless, external consent) and prompt loops without implementing the actual adapters, alongside a \RouteAuthoritySnapshot\ that binds a compiled catalog to its resolver to prevent state drift. The system also adds a three-state capability profile (\Supported\, \Unsupported\, \Unknown\) for facts like web search and image input, and provides a \ProvidersExport\ mechanism to serialize route metadata for CLI consumption.
crates/config/src/route · high confidence
Introduces durable bridge-core state management and config validation
This change adds a new \bridge-core\ library that provides a \ThreadStore\ for durable, crash-safe state persistence across bridge restarts, ensuring that chat contexts, message deduplication, and approval tokens survive process crashes. It also introduces a \validate-config\ script for both the Feishu and Telegram bridges, allowing operators to verify environment variables and filesystem permissions before starting the service. Also in this area: 4 commits (4 fixes) fixing (repo-wide).
integrations, integrations/weixin-bridge · medium confidence
Introduces secure, file-backed account session storage with cross-process locking
The \crates/secrets\ module now includes a dedicated account session storage contract (\account.rs\) that persists Codewhale account credentials and session metadata in a versioned, file-based store. This storage is protected by a shared file lock (\file\_lock.rs\) to serialize writes across processes, preventing race conditions during token refreshes or logouts. The implementation enforces strict security measures, including rejecting symlinked secret files and requiring 0600 permissions on Unix systems. Tests confirm that the system correctly handles atomic transactions, rejects insecure paths, and manages session state transitions (such as clearing companion device tokens on logout or account switch).
crates/secrets · high confidence
New /advisor command to toggle background agent watcher
Users can now control the background advisor watcher via the new /advisor command. Running /advisor on enables the watcher, which automatically runs after turns containing tool calls to provide concise LLM-based notes in the status area; /advisor off disables it, and /advisor status shows the current state. The feature is off by default and failures in the background task do not affect the parent turn.
crates/tui/src/commands/groups/core · high confidence
New /config and /import-claude commands with portable permission and status reporting
The TUI now includes a dedicated config command group that consolidates settings management, migration, and policy reporting. Users can open the canonical configuration editor via \/config\ (with optional \tui\, \web\, or \native\ arguments for muscle memory), apply bundled presets like \/config preset calm\, and manage individual settings. A new \/import-claude\ command allows users to review and apply a migration plan from Claude Code, copying standing instructions and generating a portable bundle for MCP servers, while explicitly leaving hooks and permission rules for manual review. Additionally, the \/permissions\ command provides a numbered, confirmation-gated interface to list and remove permission rules, and \/status\ displays a compact runtime report including session metrics, context window usage, and provider catalog freshness.
crates/tui/src/commands/groups/config · high confidence
New /memory and /note commands with workspace-scoped persistence
The TUI now includes a new /memory command that interacts with an authoritative native structured database for memory management, supporting operations like search, get, remember, import, export, reindex, and clear (with explicit scope and confirmation). It also introduces a /note command for managing persistent workspace notes, allowing users to add, list, show, edit, remove, and clear notes stored in a workspace-local file (preferring .codewhale/notes.md, falling back to .deepseek/notes.md). Both commands are scoped to the current workspace, with the /note command enforcing that the notes file resides within the workspace boundaries to prevent accidental modification of files outside the project context.
crates/tui/src/commands/groups/memory · high confidence
New CLI integration for DeepSeek Harness (DSH)
The TUI now includes a dedicated CLI command surface (\codewhale integrations dsh\) to manage the DeepSeek Harness integration. This addition allows users to detect the DSH environment, view detailed status reports (including connection state, bundle availability, and skin/ocean settings), and confirm installation plans. The feature establishes the foundational wiring for the DSH adapter within the Codewhale ecosystem, enabling users to connect their existing DSH setup to the Codewhale provider model.
crates/tui/src/integrations · high confidence
New TUI utility commands for attachments, automation, jobs, MCP, network, tasks, and updates
The TUI now exposes a set of slash commands under the utility group, allowing users to manage media attachments (/attach), control scheduled automations (/automation), inspect and manage background shell jobs (/jobs), configure and manage Model Context Protocol servers (/mcp), manage network allow/deny policies (/network), handle background tasks (/task), and check for or install application updates (/update) directly from the interface. These commands are registered via the portable command-contract system, with specific capabilities (workspace, media, presentation) gating access to the relevant contexts.
crates/tui/src/commands/groups/utility · high confidence
New TUI views for automations, extensions, and fleet management
The TUI now includes dedicated views for managing scheduled automations, a unified extensions inventory (covering hooks, plugins, marketplace, skills, and MCP), and a complete fleet management suite (roster, list, detail, and setup). These views provide a consistent interface for creating and editing automation schedules, browsing and filtering installed extensions, and configuring agent teams with role-based profiles and model routing.
crates/tui/src/tui/views · high confidence
New anonymous usage telemetry with durable opt-out
The \crates/telemetry\ crate introduces a new, default-on anonymous usage reporting system that collects aggregate product metrics (such as turn counts, tool calls, and error rates) while strictly excluding user content, code, or personal identifiers. The system enforces a privacy-first design: it uses a random, rotating install ID, validates all data against a closed schema to prevent data leakage, and provides a clear first-run notice. Users can permanently disable collection via a durable opt-out (config or CLI flag), which leaves a tombstone to prevent any future data emission, or temporarily disable it via environment variables without affecting stored identity.
crates/telemetry/src · high confidence
New build and catalog automation scripts
Added several Python scripts to the \scripts/\ directory to support build, branding, and catalog management. \scripts/build-lock.py\ introduces a machine-wide exclusive build lock to prevent concurrent Cargo builds from exhausting memory on shared agents. \scripts/brand/braille-mark.py\ and \scripts/brand/whale-points.py\ automate the generation of TUI launch-mark assets and the pet body point cloud from the founder's raster image, replacing hand-drawn or static assets. \scripts/catalog\_models\_dev.py\ and its associated seed files (\models\_dev\_seed.toml\, \models\_dev\_seed.lock.json\) provide an offline seed and automation for refreshing the Models.dev catalog, including credential scrubbing and drift checks. Additionally, \scripts/check-blocking-calls-budget.py\ and its JSON budget file enforce a ratchet on blocking calls in async Rust code to prevent Tokio worker parking, while \scripts/check-bundled-plugin-claims.py\ ensures bundled plugin versions in the changelog match the actual shipped assets.
scripts · high confidence
New declarative and programmatic workflow starters for release acceptance and operate scenarios
The workflows directory now includes several new starter recipes that define how agents coordinate on specific tasks. The \issue\audit.workflow.js\ file introduces a declarative, JSON-based workflow for auditing issue fixes using parallel specialist agents (code, test, docs) and synthesizing a release note. The \operate\\*.workflow.js\ files provide programmatic JavaScript starters for operational tasks: \operate\_best\_of\_n.workflow.js\ implements a best-of-N candidate search with a read-only reviewer; \operate\_parallel\_scout.workflow.js\ handles parallel scouting with partial-failure synthesis; \operate\_read\_audit.workflow.js\ performs read-only security and reliability audits; and \operate\_staged\_fix.workflow.js\ manages staged bug fixes in isolated worktrees with verification. Additionally, \stopship.workflow.js\ defines a comprehensive, read-only release acceptance fixture that verifies the Fleet, Workflow, Lane, and Runtime contracts through a strict sequence of explore, plan, review, verify, and release-lead roles, requiring explicit verdicts and source evidence at each gate.
workflows · high confidence
New execpolicy crate with layered permission rules and shell-aware deny matching
The \crates/execpolicy\ crate introduces a structured permission engine that replaces the legacy Starlark policy. It supports layered rulesets (BuiltinDefault, Agent, User) where deny rules take precedence over ask and allow actions. The engine now matches deny rules against the actual commands a shell would execute—handling wrappers like \sudo\ and \eval\, command substitutions, and redirections—rather than just the raw command string. It also includes an arity-aware prefix matching system for common tools (git, npm, cargo, docker, etc.) to allow flags without bypassing safety checks, and introduces a configurable \ApprovalMode\ (Suggest, Auto, Bypass, Never) for user-facing approval postures.
crates/execpolicy · high confidence
New localization crate for TUI strings
The \crates/localization\ crate has been introduced to manage high-visibility TUI strings, supporting 15 languages (English, Japanese, Simplified/Traditional Chinese, Brazilian Portuguese, Latin American Spanish, Vietnamese, Korean, Catalan, German, French, Indonesian, Hindi, Russian, and Ukrainian). This change provides a lightweight, bounded-initialization localization registry using \rust-i18n\ for UI chrome only, ensuring that interface text is properly translated without affecting model prompts or provider behavior.
crates/localization/src · high confidence
New packaging assets for AUR, Windows (winget), Docker, and FreeBSD
This release adds new packaging infrastructure to make CodeWhale easier to install on Linux, Windows, and macOS. On Arch Linux, a new AUR package (\codewhale-bin\) is provided with a \render.sh\ script to generate the \PKGBUILD\ and \.SRCINFO\ from verified release archives, including support for the optional compiled Bun extension host. For Windows, a new \winget\ manifest (\HunterBown.CodeWhale\) is added, supporting both NSIS and portable ZIP installers for x64 and arm64 architectures, along with a helper script to update the manifest with release checksums. A new \Dockerfile.release\ assembles container images from native musl artifacts, and documentation for FreeBSD source builds via Cargo is added.
packaging · high confidence
New progressive setup wizard with guided constitution and fleet drafting
The TUI now features a new progressive setup wizard (\/setup\) that guides users through required and optional configuration steps, including language, provider/model, trust/sandbox, and a new constitution step. Users can now have the model draft their constitution and fleet agent profiles based on guided answers, with strict privacy safeguards ensuring no secrets or file contents are leaked during drafting. The wizard also includes new cards for remote runtime configuration, tools/MCP/skills inventory, and persistence path verification, providing a comprehensive readiness summary and repair guide for new and existing users.
crates/tui/src/tui/setup · high confidence
New release management crate with throttled update checks and install-aware updates
The \crates/release\ module introduces a structured approach to release handling, featuring throttled background update checks that cache the latest release tag on disk to limit network requests to once per hour while ensuring users always see the latest notice. It includes an install-method detector that identifies whether the binary was installed via Homebrew, npm, cargo, Omarchy, or as a standalone release, ensuring that updates respect package manager ownership (e.g., suggesting \brew upgrade codewhale\ instead of in-place replacement for Homebrew installs). Additionally, it tracks the last launched version to detect upgrades and provide change notifications on first launch after an update, while also handling TLS bootstrap and platform-specific HTTP client configuration for Android.
crates/release · high confidence
New self-serve hotbar setup wizard
Users can now configure hotbar bindings through an interactive setup wizard. This new feature allows you to browse available actions by category (app, route, slash, MCP, skill, plugin), filter and search for specific commands, and assign them to hotbar slots. The wizard includes validation to ensure bindings are safe and correctly formatted, and it persists your custom configuration to disk.
crates/tui/src/tui/hotbar · high confidence
New slash commands for skills management, review, and workspace restoration
The TUI now exposes four new slash commands under the skills group: \/skills\ and \/skill\ for listing and activating skills, \/review\ to activate the review skill and send a target, and \/restore\ to list and revert the workspace to a prior snapshot. The restore command enforces a safety gate, requiring explicit trust mode or full access for non-trivial rollbacks, while the skills commands support installation, updates, and trust management via the new portable command contract.
crates/tui/src/commands/groups/skills · high confidence
New theme system with WCAG contrast enforcement and community presets
The application now includes a comprehensive color palette system that introduces new theme presets (Shoreline, Underwater, Catppuccin Mocha, Tokyo Night, Dracula, Gruvbox Dark, Claude, Matrix, Solarized Light, and Uwu) alongside the existing Whale themes. This change adds a JSON schema for user-defined theme overlays, enabling custom color configurations. It also implements automatic terminal background detection via OSC 11 queries and COLORFGBG environment variables to correctly adapt to light or dark terminal environments, preventing illegible text on mismatched backgrounds. Furthermore, the system enforces WCAG 2.x AA contrast ratios (4.5:1 for body text, 3:1 for secondary chrome) by automatically adjusting foreground colors that fail to meet these thresholds, ensuring accessibility across all themes and terminal configurations.
crates/palette · high confidence
Portable session command handlers with lifecycle and control facets
The TUI session command group has been refactored into a portable, host-agnostic implementation. Commands such as \/branch\, \/fork\, \/load\, \/new\, \/save\, \/relay\, \/remote\_control\, and \/export\ now operate through defined contract facets (\CommandSessionLifecycleContext\, \CommandSessionControlContext\, \CommandSessionExportContext\). This change isolates parsing, message composition, and action routing from the host environment, ensuring that missing capabilities fail safely with specific error messages rather than panics. The module also includes deterministic test support to verify exact message and action composition without requiring host state.
crates/tui/src/commands/groups/session · high confidence
Underwater theme introduces animated ocean column with preserved semantic styling
The Underwater theme now paints a vertical ocean column in the terminal background using a smooth, truecolor gradient that fades from surface to deep blue. This visual layer is animated with a continuous, smooth caustic shimmer effect that respects text boundaries and avoids hard cutoffs. Crucially, the implementation preserves all semantic styling—such as selection backgrounds, code highlighting, and warning colors—ensuring that user interface elements remain readable and distinct against the animated backdrop.
crates/tui/src/tui/ocean · high confidence
Unified plugin management with marketplace and local import bridges
The \/plugin\ command now provides a consolidated interface for managing plugins, introducing a marketplace system and local import bridges. Users can discover and install plugins from local catalog files via \/plugin marketplace\ (add, list, show, remove, install), where installs still require manual trust and enablement. New import bridges allow reviewing and installing plugins from DeepSeek Harness bundles (\/plugin import dsh\) and Kimi-managed plugins (\/plugin import kimi\), both routing through the standard reviewed installer to ensure security. Legacy executable tools are now listed separately under \/plugin tools\ and are clearly labeled as such, without sharing trust state with declarative bundles. The command surface is now portable, consuming workspace and presentation facets for consistent behavior.
crates/tui/src/commands/groups/plugins · high confidence
Removals
Removal of Duo, RLM, and core tool implementations
The tool system in \src/tools\ has been stripped of its core implementations and module structure. The \duo\ module (providing the Player-Coach autocoding workflow via \duo\_init\, \duo\_player\, and \duo\_coach\ tools) and the \rlm\ module (providing recursive sub-query tools like \rlm\_exec\ and \rlm\_load\) have been completely removed. Additionally, the foundational file, patch, plan, search, shell, and subagent tool implementations have been deleted, along with the \ToolRegistry\ and \ToolSpec\ trait definitions. The \mod.rs\ file no longer exports these components, effectively disabling the tool execution capabilities that were previously available in this location.
src/tools · high confidence
Removal of core engine and session management modules
The core engine implementation and its associated session management components have been removed from the \src/core\ directory. Specifically, \engine.rs\ (the main AI interaction loop and tool orchestration), \session.rs\ (conversation history and state tracking), \turn.rs\ (turn context and tool call tracking), \events.rs\ (UI event definitions), and \ops.rs\ (UI-to-engine operations) are no longer present. This eliminates the previous architecture where the CLI handled AI interactions in a background task, communicating with the UI via channels for non-blocking updates, streaming, and tool execution.
src/core · high confidence
Removal of execpolicy and sandbox modules
The \execpolicy\ and \sandbox\ modules have been removed from the codebase. This deletes the entire \src/execpolicy\ directory, including the Starlark-based policy parser, rule matching logic, and the \execpolicy check\ CLI command, as well as the \src/sandbox\ directory containing the \SandboxManager\, platform-specific sandboxing implementations (macOS Seatbelt and Linux Landlock), and the \SandboxPolicy\ definitions. Users will no longer be able to define or enforce execution policies or run commands within a restricted sandbox environment.
src/execpolicy, src/sandbox · high confidence
Removal of legacy DeepSeek client and core modules
The \src\ directory has been stripped of its previous implementation, deleting the DeepSeek HTTP client (\client.rs\), the LLM client trait and error handling (\llm\_client.rs\), context compaction logic (\compaction.rs\), command safety analysis (\command\_safety.rs\), the Duo mode state machine (\duo.rs\), the hooks system (\hooks.rs\), and the main CLI entry point (\main.rs\). This removal eliminates the existing API integration, shell safety checks, and interactive TUI features from this location.
src · high confidence
Removal of legacy slash-command subsystem
The entire legacy slash-command implementation in \src/commands\ has been removed. This deletes the command registry (\mod.rs\) and all associated modules (\config\, \core\, \debug\, \init\, \queue\, \rlm\, \session\, \skills\), eliminating commands such as \/config\, \/set\, \/yolo\, \/rlm\, \/queue\, \/init\, and \/skills\ from the application.
src/commands · high confidence
Removal of responses-api-proxy component
The \responses-api-proxy\ module and its associated \read\_api\_key\ helper have been removed from the codebase. This eliminates the local proxy server that previously forwarded requests to the DeepSeek API, including its authentication handling, CLI argument parsing, and HTTP forwarding logic.
_src/responses\_api\proxy · high confidence
Removal of the legacy DeepSeek TUI module
The entire \src/tui\ module has been removed from the codebase. This deletion eliminates the previous terminal user interface implementation, including the application state machine, tool approval overlays, clipboard handling, transcript rendering, and streaming logic. Users will no longer have access to this specific TUI interface.
src/tui · high confidence
Architecture
Extracted TUI UI modules and added fatal-signal terminal guard
The TUI UI logic has been refactored by extracting specific concerns into dedicated modules: \activity\_detail.rs\ now handles the reasoning timeline and turn inspector (Ctrl+O), \dispatch.rs\ manages message queuing and hotbar actions, \dispatch\_prepare.rs\ handles paused command logic, and \fatal\_signal\_guard.rs\ implements an async-signal-safe terminal restore for Unix aborts. This modularization cleans up the monolithic \ui.rs\ while preserving the existing user-facing behaviors for reasoning inspection, command dispatch, and crash recovery.
crates/tui/src/tui/ui · high confidence
Refactor TUI app module into focused sub-modules
The TUI application logic has been reorganized to improve maintainability and separation of concerns. The large, monolithic \app.rs\ module has been split into specialized files: \composer.rs\ now handles all text editing, cursor movement, and input history logic; \init.rs\ contains the application construction and initialization flow; \status.rs\ manages toast notifications and status messages; \types.rs\ holds plain data types and enums; and \tests.rs\ consolidates unit tests. This change does not alter user-facing behavior but cleans up the internal structure of the TUI app.
crates/tui/src/tui/app · high confidence
Behavioural changes
1 commit (0 fixes) modifying (repo-wide)
A change to existing behaviour in (repo-wide) — 1 commit, 7 files: “refactor(engine): extract the shared runtime from the TUI \[skip ci\]”.
(repo-wide) · low confidence · unverified
1 commit (0 fixes) modifying crates/app-server, crates/app-server/src/remote\_setup
A change to existing behaviour in crates/app-server, crates/app-server/src/remote\_setup — 1 commit, 4 files: “app-server: own the Runtime host and control transports; add native stats and per-call cache receipts”.
_crates/app-server, crates/app-server/src/remote\setup · medium confidence · unverified
112 commits (54 fixes) modifying crates/runtime
A change to existing behaviour in crates/runtime — 112 commits (54 fixs), 73 files: “fix(util,runtime): move the env guard beside the barrier and publish util without a runtime cycle”; “fix(tui): keep the turn metadata block out of /tree and /export”; “fix(session): refresh a metadata-only prefix instead of forking”; and 109 more (refs \#6946, \#6184, \#7251, \#6586, \#7272, …).
crates/runtime · medium confidence · unverified
12 commits (1 fix) modifying crates/tui/src/tui/views/automations
A change to existing behaviour in crates/tui/src/tui/views/automations — 12 commits (1 fix), 1 file: “refactor: extract palette into its own crate and move command\_safety to execpolicy”; “feat(tui): edit automation schedules and pin model routes”; “Carry the creating session's tool deny rules onto tasks, automations and the threads that run them”; and 9 more.
crates/tui/src/tui/views/automations · medium confidence · unverified
12 commits (4 fixes) modifying crates/tui/src/context\_report
A change to existing behaviour in crates/tui/src/context\_report — 12 commits (4 fixs), 2 files: “fix(context): headline /context with the pressure estimate the meter and gate use”; “Connect account profile constitutions to Engine turn admission”; “tests: move App fixtures to their UI owner”; and 9 more.
_crates/tui/src/context\report · medium confidence · unverified
13 commits (6 fixes) modifying crates/cli/src/update
A change to existing behaviour in crates/cli/src/update — 13 commits (6 fixs), 2 files: “fix(install): validate ownership and destinations before writes \[skip ci\]”; “fix(cli): remove the marked PATH block and Codewhale completions on uninstall”; “feat(doctor): one installation inventory; doctor --installs, --fix tidy, --restore”; and 10 more (refs \#7091).
crates/cli/src/update · medium confidence · unverified
13 commits (6 fixes) modifying crates/commands
A change to existing behaviour in crates/commands — 13 commits (6 fixs), 18 files: “fix(undo): bind file rollback to the selected recorded invocations \[skip ci\]”; “refactor(commands): move the portable command handlers into codewhale-commands”; “fix(commands): localize /undo feedback”; and 10 more.
crates/commands · medium confidence · unverified
1379 commits (569 fixes) modifying crates/tui
A change to existing behaviour in crates/tui — 1379 commits (569 fixs), 5 files: “fix(api): keep repeated hook labels distinct in plugin index”; “docs(providers): credit the Yolo-Auto contribution as unreleased”; “docs(changelog): link the provider-switch fix to its PR”; and 1362 more (refs \#6652, \#5410, \#5463, \#5361, \#5750, …).
crates/tui · medium confidence · unverified
15 commits (1 fix) modifying python
A change to existing behaviour in python — 15 commits (1 fix), 5 files: “refactor(split): codewhale-runtime crate and boundary ratchet (RS-0..RS-7) (\#6586)”; “refactor(commands): complete session group extraction boundary”; “Clean up repo for public release”; and 12 more (refs \#6586).
python · medium confidence · unverified
15 commits (6 fixes) modifying crates/skills/src
A change to existing behaviour in crates/skills/src — 15 commits (6 fixs), 12 files: “fix(workspace): keep non-home link confinement, exempt only home state”; “feat(engine): compose scoped MCP batches in the existing VM \[skip ci\]”; “feat(skills): make bundled install and discovery configurable \[skip ci\]”; and 12 more (refs \#7311).
crates/skills/src · medium confidence · unverified
16 commits (6 fixes) modifying crates/engine
A change to existing behaviour in crates/engine — 16 commits (6 fixs), 62 files: “fix(extension-host): gate dsh\_settings through BrokerContext::act and add changelog (CW-7983)”; “harness(plan,send\_later): add sub-minute timers, scheduler notify, and replan\_if (CW-8215)”; “Gate persisted records carrying tool deny rules on bumped schema versions”; and 13 more. Also in this area: 18 commits (10 fixes) fixing crates/engine; 8 commits adding/updating tests in crates/engine/tests.
(repo-wide) · medium confidence · unverified
16 commits (7 fixes) modifying crates/plugins/src/install
A change to existing behaviour in crates/plugins/src/install — 16 commits (7 fixs), 9 files: “fix(runtime): admit supported DSH packages and clarify lock refusals \[skip ci\]”; “docs: unlink intra-doc references to private items so the docs gate builds”; “fix(release): align reviewed UI admission and terminal kit \[skip ci\]”; and 13 more (refs \#2, \#3).
crates/plugins/src/install · medium confidence · unverified
17 commits (5 fixes) modifying crates/app-server-protocol
A change to existing behaviour in crates/app-server-protocol — 17 commits (5 fixs), 39 files: “feat(app-server): advertise thread\_scoped\_reads and document status\_unavailable”; “fix(app-server): return the pinned thread/list response shape”; “feat(app-server): scope git writes and add worktrees”; and 14 more.
crates/app-server-protocol · medium confidence · unverified
2 commits (1 fix) modifying crates/cli/src/metrics
A change to existing behaviour in crates/cli/src/metrics — 2 commits (1 fix), 1 file: “fix: clear the hosted clippy and Windows compile errors on main”; “app-server: own the Runtime host and control transports; add native stats and per-call cache receipts”.
crates/cli/src/metrics · medium confidence · unverified
2 commits (1 fix) modifying crates/engine/src/compaction
A change to existing behaviour in crates/engine/src/compaction — 2 commits (1 fix), 5 files: “fix(engine): stop replaying inference with unknown outcomes \[skip ci\]”; “refactor(engine): extract the shared runtime from the TUI \[skip ci\]”.
crates/engine/src/compaction · medium confidence · unverified
2 commits (1 fix) modifying crates/engine/src/receipts
A change to existing behaviour in crates/engine/src/receipts — 2 commits (1 fix), 2 files: “feat(receipts): carry token counts and the /cost line; count saved-session usage in metrics (CW-8032)”; “fix(receipts): a thread receipt carries tokens only, with no provider-call count (CW-8032)”.
crates/engine/src/receipts · medium confidence · unverified
2 commits (1 fix) modifying crates/memory
A change to existing behaviour in crates/memory — 2 commits (1 fix), 25 files: “fix(secrets): work on a mount without Unix mode bits or advisory locks”; “Vendor codewhale-memory 0.2.0 as crates/memory”.
crates/memory · medium confidence · unverified
2 commits (1 fix) modifying extensions/vscode/scripts
A change to existing behaviour in extensions/vscode/scripts — 2 commits (1 fix), 1 file: “Admit one reviewed dev-only advisory in the extension packager”; “fix(vscode): patch MSAL packager dependency and remove waiver \[skip ci\]”.
extensions/vscode/scripts · medium confidence · unverified
2 commits (1 fix) modifying integrations/dingtalk-bridge
A change to existing behaviour in integrations/dingtalk-bridge — 2 commits (1 fix), 2 files: “fix(bridges): bound Runtime events and preserve interrupted turns \[skip ci\]”; “feat(integrations): reuse Engine reviews and durable chat admission \[skip ci\]”.
integrations/dingtalk-bridge · medium confidence · unverified
2 commits (1 fix) modifying patches/crossterm-0.29.0
A change to existing behaviour in patches/crossterm-0.29.0 — 2 commits (1 fix), 47 files: “feat: integrate the verified 0.10.3 combined Engine batch \[skip ci\]”; “fix(release): keep emoji in Windows Terminal pastes and make reviewed host plugins survive restart”.
patches/crossterm-0.29.0 · medium confidence · unverified
20 commits (6 fixes) modifying crates/tui/src/lib
A change to existing behaviour in crates/tui/src/lib — 20 commits (6 fixs), 3 files: “test(fleet): move the last fleet -\> lib.rs test edge to the CLI tests”; “feat(CW-7503): intent, evidence tiers, a refutation pass, changed contracts and the summary shape”; “fix(deepseek): Responses reasoning placeholder, 128K cap at max effort, exec finish\_reason; localize /statusline hints”; and 17 more (refs \#6939).
crates/tui/src/lib · medium confidence · unverified
2331 commits (819 fixes) modifying (repo-wide)
A change to existing behaviour in (repo-wide) — 2331 commits (819 fixs), 24 files: “fix(api): keep repeated hook labels distinct in plugin index”; “wip: checkpoint v0.8.67 fallback candidate”; “docs(providers): credit the Yolo-Auto contribution as unreleased”; and 2311 more (refs \#561, \#562, \#563, \#564, \#567, …).
(repo-wide) · low confidence · unverified
27 commits (9 fixes) modifying vendor/codewhale-ratatui
A change to existing behaviour in vendor/codewhale-ratatui — 27 commits (9 fixs), 110 files: “style(release): format avatar routes and vendored module declarations”; “feat(tui): add a live terminal work dock”; “Composer: grow past the density cap for long drafts”; and 24 more.
vendor/codewhale-ratatui · medium confidence · unverified
3 commits (0 fixes) modifying crates/engine/src/image\_attach
A change to existing behaviour in crates/engine/src/image\_attach — 3 commits, 2 files: “refactor(protocol): share Runtime wire contracts and generated SDK types \[skip ci\]”; “feat: drive Responses behavior from a dialect profile (\#7261)”; “refactor(engine): extract the shared runtime from the TUI \[skip ci\]” (refs \#7261).
_crates/engine/src/image\attach · medium confidence · unverified
3 commits (0 fixes) modifying crates/engine/src/project\_context
A change to existing behaviour in crates/engine/src/project\_context — 3 commits, 5 files: “chore: remove unreferenced code and stale dead\_code allows (\#7272)”; “feat(engine): preserve native watch, MCP policy and local Soul ownership”; “refactor(engine): extract the shared runtime from the TUI \[skip ci\]” (refs \#7272).
_crates/engine/src/project\context · medium confidence · unverified
3 commits (0 fixes) modifying crates/engine/src/work\_graph
A change to existing behaviour in crates/engine/src/work\_graph — 3 commits, 12 files: “harness(plan,send\_later): add sub-minute timers, scheduler notify, and replan\_if (CW-8215)”; “refactor: retire unused APIs and the duplicate agent catalog”; “refactor(engine): extract the shared runtime from the TUI \[skip ci\]”.
_crates/engine/src/work\graph · medium confidence · unverified
31 commits (14 fixes) modifying crates/plugins/src
A change to existing behaviour in crates/plugins/src — 31 commits (14 fixs), 27 files: “Clarify unavailable plugin snapshot diagnostics \[skip ci\]”; “fix(plugins): give the finalized stage root one non-inheriting ACL”; “fix(plugins): the agent's review is decided on the record at write time, and an agent's record belongs to its project (CW-8053)”; and 28 more (refs \#2, \#3, \#7261).
crates/plugins/src · medium confidence · unverified
31 commits (6 fixes) modifying vendor
A change to existing behaviour in vendor — 31 commits (6 fixs), 114 files: “style(avatars): keep shared pack helpers and tests lint clean”; “Composer: grow past the density cap for long drafts”; “pet: one pet mode that switches views, keys that reach a Mac, a crisp live whale”; and 28 more (refs \#6968).
vendor · medium confidence · unverified
35 commits (6 fixes) modifying crates/tui/src/conformance
A change to existing behaviour in crates/tui/src/conformance — 35 commits (6 fixs), 8 files: “test(conformance): order a parallel batch's starts and completions canonically”; “feat(engine): fence side-browser calls to the account that admitted the turn”; “fix(session): recover the crash checkpoint on every attach; settle two conformance goldens”; and 31 more (refs \#7261).
crates/tui/src/conformance · medium confidence · unverified
4 commits (0 fixes) modifying crates/engine/src/extension\_host/protocol
A change to existing behaviour in crates/engine/src/extension\_host/protocol — 4 commits, 1 file: “feat(extension-host): one ticketed dsh/broker envelope and refusing broker table (CW-8012)”; “feat(extension-host): gate service acts and retain sealed act receipts”; “refactor(protocol): share Runtime wire contracts and generated SDK types \[skip ci\]”; and 1 more.
_crates/engine/src/extension\host/protocol · medium confidence · unverified
4 commits (1 fix) modifying crates/account-cli
A change to existing behaviour in crates/account-cli — 4 commits (1 fix), 8 files: “fix(release): clear the 0.10.3 documentation and lint gates”; “refactor(split): move the cloud account CLI into codewhale-account-cli (S50, CW-6129) \[skip ci\]”; “feat(runtime): reconcile reviewed plugins and unify native account setup \[skip ci\]”; and 1 more.
crates/account-cli · medium confidence · unverified
4 commits (2 fixes) modifying crates/engine/src/hooks
A change to existing behaviour in crates/engine/src/hooks — 4 commits (2 fixs), 3 files: “mcp: project servers load only from approved project mcp.json contents”; “fix(engine): lean follow-ons — retraction, surface pins (CW-7923, CW-7926)”; “fix(engine): keep injected-route config when preflight fails \[skip ci\]”; and 1 more.
crates/engine/src/hooks · medium confidence · unverified
4 commits (2 fixes) modifying crates/engine/src/vision
A change to existing behaviour in crates/engine/src/vision — 4 commits (2 fixs), 2 files: “feat(transport): allow an optional host policy on model HTTP clients (\#7251)”; “fix(net): go direct when a local proxy disappears”; “refactor(engine): extract the shared runtime from the TUI \[skip ci\]”; and 1 more (refs \#7251).
crates/engine/src/vision · medium confidence · unverified
4 commits (2 fixes) modifying crates/lsp
A change to existing behaviour in crates/lsp — 4 commits (2 fixs), 4 files: “test(lsp): give the closed-channel diagnostics test a deadline it cannot miss (CW-7875)”; “fix(mcp,lsp): answer server requests instead of taking them as replies (\#7319)”; “refactor(split): extract dependencies, snapshot, lsp and repl crates (S15, CW-6129) \[skip ci\]”; and 1 more (refs \#7319).
crates/lsp · medium confidence · unverified
4 commits (2 fixes) modifying crates/repl
A change to existing behaviour in crates/repl — 4 commits (2 fixs), 3 files: “feat(plugins): trust, enable, disable and revoke plugins from a shell, recorded by who approved (CW-7891)”; “fix(repl): truncate stdout by bytes, not chars (\#7296)”; “fix(runtime): drain noisy REPL stderr and bound telemetry buffers \[skip ci\]”; and 1 more (refs \#7296).
crates/repl · medium confidence · unverified
5 commits (0 fixes) modifying crates/sanitize, crates/workflow
A change to existing behaviour in crates/sanitize, crates/workflow — 5 commits, 2 files: “Handle plural token names, prefixed PEM blocks, JSON escapes, and compact members”; “Scan the rest of a key line, URL userinfo and flag values for credentials”; “Share one credential detector between the sanitizer and the workflow redactor”; and 2 more. Also in this area: 1 commit (1 fix) fixing crates/sanitize; New sanitize crate for centralized secret redaction and output cleaning.
crates/sanitize · medium confidence · unverified
5 commits (0 fixes) modifying crates/tui/src/conformance/mcp
A change to existing behaviour in crates/tui/src/conformance/mcp — 5 commits, 2 files: “feat: integrate the verified 0.10.3 combined Engine batch \[skip ci\]”; “test(conformance): name the missed spelling instead of printing the token”; “test(conformance): widen the recorded MCP parity suite from 2 cases to 15”; and 2 more.
crates/tui/src/conformance/mcp · medium confidence · unverified
5 commits (1 fix) modifying crates/engine/src/core/engine/tool\_catalog
A change to existing behaviour in crates/engine/src/core/engine/tool\_catalog — 5 commits (1 fix), 1 file: “feat(engine): lean eager tool surface, terse schemas (CW-7926)”; “fix(engine): keep injected-route config when preflight fails \[skip ci\]”; “feat(modes): retire Operate as a mode”; and 2 more.
_crates/engine/src/core/engine/tool\catalog · medium confidence · unverified
5 commits (1 fix) modifying crates/tui/src/config\_stay\_tests
A change to existing behaviour in crates/tui/src/config\_stay\_tests — 5 commits (1 fix), 7 files: “feat: integrate the verified 0.10.3 combined Engine batch \[skip ci\]”; “refactor(split): S19 codewhale-plugins crate (CW-6129) \[skip ci\]”; “refactor(split): provider pre-cuts into codewhale-models and runtime (S22, CW-6129) \[skip ci\]”; and 2 more.
_crates/tui/src/config\_stay\tests · medium confidence · unverified
5 commits (2 fixes) modifying crates/app-server/src/acp\_server
A change to existing behaviour in crates/app-server/src/acp\_server — 5 commits (2 fixs), 2 files: “refactor(acp): serve ACP on the official SDK agent connection (CW-8043 ACP-SDK-S1)”; “fix(acp): a plugin tool's nested approval reaches the editor as a permission request that names the nested tool (CW-8050)”; “feat(engine): recover external-agent sessions and durable approval responders”; and 2 more.
_crates/app-server/src/acp\server · medium confidence · unverified
537 commits (244 fixes) modifying crates/tui/src
A change to existing behaviour in crates/tui/src — 537 commits (244 fixs), 15 files: “fix(security): redact tool output as it enters the transcript; doctor and scrub for stored secrets”; “fix(tui): resume and fork route to exec without a terminal (CW-7891)”; “feat(doctor): probe search provider reachability”; and 534 more (refs \#6511, \#6510, \#5750, \#10, \#21, …).
crates/tui/src · medium confidence · unverified
54 commits (27 fixes) modifying crates/engine/src/runtime\_threads
A change to existing behaviour in crates/engine/src/runtime\_threads — 54 commits (27 fixs), 4 files: “fix: unblock contributor PR Rust validation \[skip ci\]”; “style: cargo fmt --all”; “Gate persisted records carrying tool deny rules on bumped schema versions”; and 51 more (refs \#7289).
_crates/engine/src/runtime\threads · medium confidence · unverified
59 commits (25 fixes) modifying crates/app-server/src/runtime\_api
A change to existing behaviour in crates/app-server/src/runtime\_api — 59 commits (25 fixs), 44 files: “fix(api): keep repeated hook labels distinct in plugin index”; “feat(app-server): resolve git, diff and file reads in a thread's own workspace”; “fix(engine,app-server): find Windows agent shims and keep probe refusals truthful (CW-7879)”; and 56 more (refs \#7277, \#7310, \#2, \#3).
_crates/app-server/src/runtime\api · medium confidence · unverified
6 commits (0 fixes) modifying crates/engine/src/prompts
A change to existing behaviour in crates/engine/src/prompts — 6 commits, 2 files: “feat(engine): ask the user with options when a choice is theirs”; “feat(engine): slim prompt core, no skills index, registry guidance in tool descriptions (CW-7924)”; “feat(engine): eager sub-agents and goals by default, with clearer tool text”; and 3 more.
crates/engine/src/prompts · medium confidence · unverified
6 commits (1 fix) modifying crates/engine/src/core/authority
A change to existing behaviour in crates/engine/src/core/authority — 6 commits (1 fix), 5 files: “fix(engine): collapse four needless\_bool tails flagged by clippy 1.100”; “chore: remove unreferenced code and stale dead\_code allows (\#7272)”; “feat(plugins): trust, enable, disable and revoke plugins from a shell, recorded by who approved (CW-7891)”; and 3 more (refs \#7272).
crates/engine/src/core/authority · medium confidence · unverified
6 commits (1 fix) modifying crates/plugins/src/marketplace/parsers
A change to existing behaviour in crates/plugins/src/marketplace/parsers — 6 commits (1 fix), 6 files: “Resolve portable marketplace sources in their publisher context \[skip ci\]”; “refactor(split): S19 codewhale-plugins crate (CW-6129) \[skip ci\]”; “feat(plugins): read DeepSeek Harness plugin market indexes (CW-7051)”; and 3 more.
crates/plugins/src/marketplace/parsers · medium confidence · unverified
6 commits (3 fixes) modifying crates/app-server/src/thread\_control
A change to existing behaviour in crates/app-server/src/thread\_control — 6 commits (3 fixs), 1 file: “Fix API compatibility admission and sandboxed PTY sessions for 0.10.2 \[skip ci\]”; “fix(app-server): reject ignored list and read parameters \[skip ci\]”; “fix: qualify 0.10.1 Engine and TypeScript integration”; and 3 more.
_crates/app-server/src/thread\control · medium confidence · unverified
6 commits (3 fixes) modifying crates/engine/src/client/responses
A change to existing behaviour in crates/engine/src/client/responses — 6 commits (3 fixs), 1 file: “fix: address rate limit review findings”; “feat: add bounded HTTP rate-limit recovery”; “fix(engine): withhold tools whose encoded name exceeds the route limit; make decode exact (CW-8041)”; and 3 more (refs \#7261).
crates/engine/src/client/responses · medium confidence · unverified
7 commits (1 fix) modifying crates/agent-contract
A change to existing behaviour in crates/agent-contract — 7 commits (1 fix), 3 files: “refactor(split): move canonical\_action and tool contract helpers out of tui (S29/S30, CW-6129) \[skip ci\]”; “refactor(config): consume shared home and approval owners \[skip ci\]”; “fix(split): restore Windows builds and close review findings on the 0.10.3 crate split”; and 4 more.
crates/agent-contract · medium confidence · unverified
7 commits (2 fixes) modifying crates/sandbox
A change to existing behaviour in crates/sandbox — 7 commits (2 fixs), 11 files: “fix(sandbox): protect git hooks, git config and .codewhale from sandboxed writes”; “fix(clippy): clear utility lint errors”; “feat(sandbox): run tool commands as another user through a managed-only launcher (CW-8021)”; and 4 more.
crates/sandbox · medium confidence · unverified
7 commits (3 fixes) modifying crates/net
A change to existing behaviour in crates/net — 7 commits (3 fixs), 5 files: “feat: integrate the verified 0.10.3 combined Engine batch \[skip ci\]”; “fix(engine): revert and private-host fetch follow the user's own choices”; “net,cli,engine,docs: harden private-network rules, IPv6 embedded forms, redirect hops and bundle guards”; and 4 more.
crates/net · medium confidence · unverified
7 commits (3 fixes) modifying crates/snapshot
A change to existing behaviour in crates/snapshot — 7 commits (3 fixs), 5 files: “docs: unlink intra-doc references to private items so the docs gate builds”; “fix(state,snapshot): keep index entries after a torn line and clamp prune age (\#7316)”; “fix(snapshot): match recorded workspace paths case-insensitively on Windows”; and 4 more (refs \#7316, \#7317).
crates/snapshot · medium confidence · unverified
79 commits (35 fixes) modifying crates/tui/src/tui/pet\_watch
A change to existing behaviour in crates/tui/src/tui/pet\_watch — 79 commits (35 fixs), 16 files: “feat(pet): the preview work tour visits every glyph, 工具 included”; “fix(tui): preserve concrete pet activity and lint-safe painting”; “Let the entire dot whale become a readable living character”; and 75 more (refs \#6153, \#6224, \#7097).
_crates/tui/src/tui/pet\watch · medium confidence · unverified
8 commits (3 fixes) modifying crates/engine/src/runtime\_threads/manager
A change to existing behaviour in crates/engine/src/runtime\_threads/manager — 8 commits (3 fixs), 2 files: “feat(engine,app-server): ACP options route, typed control refusals and agent-reported usage (CW-7879 EA-04)”; “fix(engine): stop an abandoned external-agent startup inside the shutdown barrier (CW-7879)”; “fix(engine): answer a pending permission request as cancelled on interrupt, send additionalDirectories when advertised, project plan/mode/command updates (CW-7958 REF-09)”; and 5 more.
_crates/engine/src/runtime\threads/manager · medium confidence · unverified
8 commits (3 fixes) modifying crates/skill-install
A change to existing behaviour in crates/skill-install — 8 commits (3 fixs), 9 files: “docs: unlink intra-doc references to private items so the docs gate builds”; “docs(installers): add neutral CHANGELOG note and cap note on download\_with\_cap”; “fix(skills): detect the archive prefix from every entry and import hidden files (\#7311)”; and 5 more (refs \#7311).
crates/skill-install · medium confidence · unverified
9 commits (1 fix) modifying crates/plugins/src/marketplace
A change to existing behaviour in crates/plugins/src/marketplace — 9 commits (1 fix), 5 files: “docs: unlink intra-doc references to private items so the docs gate builds”; “feat(plugins): refresh the first-party catalog snapshot per release; replace only the embedded copy (CW-8185)”; “test(tui,plugins): regenerate conformance goldens and update marketplace/exec checks for the 0.10.3 changes”; and 6 more.
crates/plugins/src/marketplace · medium confidence · unverified
Catalog authority shifts to operator-declared routes and reviewed intrinsic facts
The catalog configuration now relies on explicit, validated operator-declared model routes (ConfiguredModel) and a reviewed intrinsic catalog (ReviewedCatalog) rather than compiled model lists. This change introduces strict validation for custom model entries, ensuring that provider, base\_url, and id fields are well-formed and unique, while preserving exact endpoint model declarations through the runtime. Additionally, it integrates Codewhale corrections (CatalogCorrections) to patch Models.dev rows with field-level updates for pricing, limits, and reasoning controls, ensuring these corrections hold both online and offline. The system now supports namespaced model-only catalog entries and retains map identities for provider precedence, providing a more robust and predictable catalog resolution mechanism.
crates/config/src/catalog · high confidence
Core engine refactoring: session tree, context references, and tool parsing
The core engine has been restructured to support durable session trees and more robust context handling. A new session model splits the previous monolithic session into a persistent Thread (with an append-only journal and leaf-cursor for branching/forking) and an ephemeral Session (per-turn state). Context references and media attachments are now parsed and persisted independently of the UI, with improved path extraction for attached files. The legacy text-based tool-call parser has been moved into core, updated to use the new codewhale XML markers, and hardened to reject malformed arguments. Additionally, a prefix-cache stability manager has been added to detect system prompt or tool-list drift, and background jobs now include detailed retry and history tracking.
crates/core/src · high confidence
Doctor diagnostics now run offline by default with enhanced privacy and local provider support
The doctor command now performs diagnostics offline by default, meaning it will not check for updates, probe APIs, or test search providers unless explicitly requested, ensuring user privacy and reducing unnecessary network traffic. This change includes redacting sensitive information like plugin review URLs and credential-shaped values from plain-text config files in diagnostic reports. Additionally, the doctor now recognizes and supports first-class local DS4 (OpenAI-compatible) provider setups, allowing users to verify local model connectivity without external dependencies. The update reporting system has also been improved to canonicalize release tags and correctly identify package managers for managed installs, providing clearer and more accurate update instructions.
crates/tui/src/doctor · high confidence
Explicit tool execution outcomes and parallel-safe scheduling
The tools crate now exposes explicit terminal states (Succeeded, Failed, Denied, InvalidArguments, Cancelled, TimedOut) via the new ToolExecutionOutcome type, allowing the runtime to distinguish cancellations and permission denials from generic failures without relying on legacy result text. It also introduces a prepared-call seam (PreparedToolCall) that records resource claims (file reads/writes, terminals, global exclusives) before execution, enabling the scheduler to batch non-conflicting tool calls in parallel while preserving order for conflicting ones.
crates/tools · high confidence
Introduce typed debug observation contracts for receipts, undo, diffs, and diagnostics
The debug command subsystem now exposes structured, portable data shapes instead of opaque or host-bound objects. New facet modules define typed contracts for debug receipts (including approval deciders and action totals), conversation undo outcomes (restored files, sync payloads, and error states), diff observations (git patches and baseline status), and diagnostics (prompt source maps, tool projections, and turn stop diagnostics). These changes allow the TUI and other portable renderers to consume consistent, schema-validated debug data while keeping mutable state and host-specific logic in the engine.
crates/command-contract/src/facets · high confidence
Localization build script resolves manifest directory at runtime
The build script for the localization crate now explicitly resolves the package's manifest directory using the CARGO\_MANIFEST\_DIR environment variable at execution time. This change ensures that the build process correctly locates the 'locales' directory and generates the necessary i18n data files, even when Cargo reuses build artifacts across different worktrees sharing a target directory. This improves the reliability of the localization setup during compilation.
crates/localization · high confidence
Migrate config policy commands to portable command contract
The \/permissions\ and \/status\ commands in the config group are now registered through the portable command contract, enabling them to operate via capability-driven facets rather than direct \App\ access. This migration preserves existing behavior: \/permissions\ continues to list, remove, and manage tool ask rules with workspace scoping and confirmation previews, while \/status\ reports runtime state including session pins, fleet drift, and token usage without rewriting config files or masking sensitive data. Host-specific result mapping ensures actions like \PermissionRulesChanged\ are correctly translated to TUI updates, and comprehensive tests verify that malformed configs, missing files, and optional config drift are handled gracefully without side effects.
crates/tui/src/commands · high confidence
New TUI widget library with legacy compatibility and acceptance tests
The TUI widget layer has been restructured into a new, modular crate under \crates/tui/src/tui/widgets\. This introduces dedicated components for the chat transcript (\ChatWidget\), sub-agent status tracking (\DelegateCard\ and \FanoutCard\), inline approval bands (\ApprovalWidget\), and a pending-input preview system (\PendingInputPreview\). To ensure a safe migration, the previous rendering logic has been preserved as frozen legacy counterparts (e.g., \legacy\_composer\, \legacy\_transcript\) and is now used as the source of truth for comprehensive acceptance tests that verify the new widgets render identically across various locales, themes, and terminal dimensions.
crates/tui/src/tui/widgets · high confidence
New first-run onboarding flow with language, provider, and trust steps
The TUI now presents a structured, calm onboarding experience for first-time users, replacing the previous ad-hoc setup surfaces. This flow guides users through three required decisions: selecting a UI language (with an auto-detect option and support for locales like Korean, Japanese, and Simplified Chinese), choosing an AI provider or model route, and explicitly granting or denying workspace trust. The language picker persists the user's choice immediately, while the trust screen ensures security boundaries are respected by requiring explicit key presses (1/Y, 2/U, 3/N) rather than allowing reflexive acceptance. The flow concludes with a 'Ready' screen that hands the user directly to the composer pre-seeded with a first task, avoiding further educational interruptions.
crates/tui/src/tui/onboarding · high confidence
New model metadata constants and stop-reason logic in the models crate
The \crates/models\ crate now defines specific context window and output token limits for DeepSeek V4 and Kimi K3 models, establishing conservative baselines and documented maximums for these providers. It also introduces logic to accurately detect when a response is cut off due to output ceilings or other incomplete states, ensuring that stop reasons like 'length' or 'max\_tokens' are correctly interpreted even when usage data contradicts the provider's stop signal. Additionally, the crate re-exports core message types and introduces new structures for container metadata and server tool usage counters to support code-execution workflows. Also in this area: 1 commit (0 fixes) modifying crates/models; 1 commit (1 fix) fixing crates/models.
crates/models · low confidence · unverified
Portable command contracts for TUI configuration, policy, navigation, and diagnostics
The TUI now exposes its internal command logic through a set of portable host adapters (config, policy, navigation, and debug/diagnostics) that bridge the application state to a stable command contract. Users benefit from consistent, cross-platform behavior in settings management, permission rules, session status, and debug operations, as the TUI now delegates to these adapters rather than handling commands directly. This change ensures that configuration changes, policy observations, and diagnostic data are projected through a unified interface, improving reliability and testability across different operating systems. Also in this area: 7 commits (6 fixes) fixing crates/tui/src; 3 commits (1 fix) modifying crates/tui/src/commands.
crates/tui/src/commands/contract · medium confidence · unverified
Project commands now use the portable command contract
The \/goal\, \/init\, \/lsp\, and \/share\ commands in the TUI have been refactored to use the new portable command contract (FEAT-021). This change replaces the previous command registration mechanism with a unified, capability-based dispatch system, ensuring consistent behavior and easier maintenance across project-related features. The \/goal\ command now supports context-dependent goal declaration, \/init\ generates AGENTS.md with rich project context, \/lsp\ manages LSP integration status, and \/share\ allows redacted session sharing as a secret gist.
crates/tui/src/commands/groups/project · high confidence
Redesigned approval interface with sandbox elevation and persistent rules
The approval system has been refactored into distinct modules to improve clarity and security. Users now see a sandbox elevation modal when a tool is denied, offering options to allow network access, grant specific write permissions, or abort, with the default selection safely set to Abort to prevent accidental privilege escalation. Approval cards now display human-readable previews of persistent permission rules (e.g., "run cargo test") instead of raw configuration syntax, and include a bounded preview of file changes or shell commands. The system also introduces a timeout mechanism for interactive tool approvals, which fail-closed (denies) if no decision is made within the configured duration, and supports persistent allow rules that are saved with a visible preview to ensure users understand what they are permitting.
crates/tui/src/tui/approval · high confidence
Redesigned work surface with responsive strip and dock tabs
The work surface rendering has been rewritten to support a responsive strip that auto-opens only for active work (agents, to-dos, jobs) rather than persisting for all panels, and includes a dock tab row for navigation. The layout now adapts to terminal width and height, falling back to a top strip for narrow screens, and displays a goal title with a to-do progress receipt above the scrollable row list. Sub-agent rows degrade gracefully on narrow widths, dropping token counts and identity columns while preserving the objective, and the new implementation includes comprehensive tests to ensure visual and interaction parity with the previous renderer.
_crates/tui/src/tui/work\surface/render · high confidence
Redirect worker now enforces HTTPS for codewhale.net
The web/redirect worker has been updated to ensure all redirects to codewhale.net use the HTTPS protocol. Previously, redirects might have allowed HTTP; the new implementation explicitly sets the protocol to https: before performing the 301 redirect, improving security for users being redirected to the site.
web/redirect · high confidence
Restructured transcript history rendering with new cell types and property-based tests
The history rendering module has been refactored into a modular structure with dedicated files for specific transcript cell types, including agent activity, archived context, automation receipts, checklists, file mutations, LaTeX math, messages, and plans. This change introduces new visual components such as compact agent activity cards, structured automation receipts with lifecycle states (fired, started, completed, failed, etc.), detailed file mutation receipts showing created/updated/deleted/renamed files with diffs, and LaTeX math rendering via Unicode substitution. The rendering now uses property-based tests instead of brittle glyph assertions, ensuring visual consistency while allowing design changes. Shared constants for tool card budgets and limits have been centralized, and the system now supports more granular control over how different types of tool outputs and metadata are displayed in the transcript.
crates/tui/src/tui/history · high confidence
Rust port of the Codewhale pet simulation and braille rendering
The ambient life TUI now uses a Rust implementation of the pet simulation core (pet\_sim.rs) and a shared braille renderer (pet\_widget.rs) to display animated creatures like whales and jellyfish. This change replaces the previous ASCII-based silhouettes with dot-braille poses defined in native\_poses.rs and whale-points.tsv, ensuring deterministic, frame-accurate animations that match the existing TypeScript/Swift engine. The new system supports multiple creature archetypes (e.g., Gyre, Strike, Pod) and channels (e.g., Reasoning, Tool, Agent), rendering them as ambient indicators of system activity. It also includes a 'cameo' feature that plays a short, deterministic whale animation upon task completion, with logic to avoid overlapping text and handle narrow viewports gracefully.
_crates/tui/src/tui/ambient\life · high confidence
Structured motion policy with coalesced frame scheduling
The TUI now enforces a unified motion policy (Full, Reduced, Still) that controls decorative animations, status spinners, and streaming display cadence. A new coalesced frame-request scheduler merges multiple animation requests into a single wake deadline, preventing redundant redraws while respecting the main UI poll loop. In Reduced mode, decorative motion and catch-up bursts are disabled in favor of static calm markers, while streaming continues at the steady display clock rather than slowing down. In Still mode, only state-change redraws occur, with no animation frames requested. This ensures that terminal compatibility limits (like frame caps) do not masquerade as accessibility preferences or disable authored motion unnecessarily.
crates/tui/src/tui/motion · high confidence
Fixes
1 commit (1 fix) fixing crates/engine/src/extension\_host/dsh\_remote
A fix in crates/engine/src/extension\_host/dsh\_remote — 1 commit (1 fix), 1 file: “fix(extension-host): harden the remote broker arm for reviewed plugins only (CW-8000)”.
_crates/engine/src/extension\_host/dsh\remote · medium confidence · unverified
107 commits (57 fixes) fixing crates/engine/src
A fix in crates/engine/src — 107 commits (57 fixs), 68 files: “fix: unblock contributor PR Rust validation \[skip ci\]”; “style: cargo fmt --all”; “harness(plan,send\_later): add sub-minute timers, scheduler notify, and replan\_if (CW-8215)”; and 104 more (refs \#6184, \#7251, \#7277, \#7319, \#7272, …).
crates/engine/src · medium confidence · unverified
109 commits (67 fixes) fixing crates/engine/src/tools
A fix in crates/engine/src/tools — 109 commits (67 fixs), 151 files: “fix: unblock contributor PR Rust validation \[skip ci\]”; “style: cargo fmt --all”; “fix(subagent): only scan nested repositories at the worktree anchor”; and 106 more (refs \#6553, \#7272, \#7276, \#7291, \#7254, …).
crates/engine/src/tools · medium confidence · unverified
11 commits (7 fixes) fixing crates/engine/src/mcp
A fix in crates/engine/src/mcp — 11 commits (7 fixs), 12 files: “fix: clear the hosted clippy and Windows compile errors on main”; “fix(mcp,lsp): answer server requests instead of taking them as replies (\#7319)”; “fix: gate Unix-only MCP test constructor on Unix”; and 8 more (refs \#7319).
crates/engine/src/mcp · medium confidence · unverified
12 commits (7 fixes) fixing crates/engine/src/core/engine/turn\_loop
A fix in crates/engine/src/core/engine/turn\_loop — 12 commits (7 fixs), 5 files: “fix: clear the hosted clippy and Windows compile errors on main”; “feat(engine): compose scoped MCP batches in the existing VM \[skip ci\]”; “fix(engine): send the goal objective once per live context window \[skip ci\]”; and 9 more.
_crates/engine/src/core/engine/turn\loop · medium confidence · unverified
12 commits (8 fixes) fixing crates/engine/src/llm\_client
A fix in crates/engine/src/llm\_client — 12 commits (8 fixs), 5 files: “fix(release): clear the 0.10.3 documentation and lint gates”; “fix: address rate limit review findings”; “fix(llm): print quota-window reset as plain text (CW-8209)”; and 9 more (refs \#7275, \#7261, \#7253).
_crates/engine/src/llm\client · medium confidence · unverified
14 commits (8 fixes) fixing crates/tui/src/tui/persistence\_actor
A fix in crates/tui/src/tui/persistence\_actor — 14 commits (8 fixs), 1 file: “test(tui): harden persistence backlog ratchet”; “fix(models): deepseek-flash reasons — stop leaking its reasoning into prose”; “fix(session): fence publications to captured native sources \[skip ci\]”; and 11 more (refs \#6212).
_crates/tui/src/tui/persistence\actor · medium confidence · unverified
15 commits (8 fixes) fixing crates/util
A fix in crates/util — 15 commits (8 fixs), 12 files: “refactor(acp): add the official ACP SDK dependency and share the bounded line reader (ACP-SDK-S0)”; “fix(util,runtime): move the env guard beside the barrier and publish util without a runtime cycle”; “fix(util): drop the trailing zero from context window labels (\#7295)”; and 12 more (refs \#7295, \#7091).
crates/util · medium confidence · unverified
17 commits (10 fixes) fixing integrations/wecom-bridge
A fix in integrations/wecom-bridge — 17 commits (10 fixs), 2 files: “feat(bridge): support natural-language approval responses”; “feat(integrations): add WeCom (企业微信) intelligent robot bridge”; “fix(bridges): bound Runtime events and preserve interrupted turns \[skip ci\]”; and 14 more (refs \#3468).
integrations/wecom-bridge · medium confidence · unverified
181 commits (91 fixes) fixing crates/app-server/src
A fix in crates/app-server/src — 181 commits (91 fixs), 23 files: “refactor(acp): serve ACP on the official SDK agent connection (CW-8043 ACP-SDK-S1)”; “refactor(acp): add the official ACP SDK dependency and share the bounded line reader (ACP-SDK-S0)”; “feat(acp): attach the extension host and configured MCP servers to ACP sessions (CW-7941)”; and 176 more (refs \#5749, \#6246, \#7277, \#6513, \#6553, …).
crates/app-server/src · medium confidence · unverified
24 commits (15 fixes) fixing crates/engine/src/client
A fix in crates/engine/src/client — 24 commits (15 fixs), 24 files: “fix(providers): refresh the Runcrate public roster without a key \[skip ci\]”; “fix(engine): Anthropic tool-name encoding, usage-journal headroom, pre-turn watchdog, lenient session list”; “fix: clear the hosted clippy and Windows compile errors on main”; and 21 more (refs \#7275, \#7261).
crates/engine/src/client · medium confidence · unverified
2537 commits (1350 fixes) fixing crates/tui/src/tui
A fix in crates/tui/src/tui — 2537 commits (1350 fixs), 98 files: “feat(work-graph): lifecycle bindings + stale truth”; “test: align macos CI assertions”; “feat(tui): support transcript scrollbar dragging”; and 2507 more (refs \#5550, \#5243, \#6652, \#6084, \#3032, …).
crates/tui/src/tui · medium confidence · unverified
27 commits (24 fixes) fixing crates/tui-kit/src
A fix in crates/tui-kit/src — 27 commits (24 fixs), 17 files: “fix: unblock contributor PR Rust validation \[skip ci\]”; “fix(release): clear the 0.10.3 documentation and lint gates”; “fix(tui): parse a table cell's markup before wrapping it”; and 24 more (refs \#7144, \#7273).
crates/tui-kit/src · medium confidence · unverified
3 commits (2 fixes) fixing crates/dependencies
A fix in crates/dependencies — 3 commits (2 fixs), 1 file: “fix(app-server): neutralize repository-defined git filter drivers on status reads (P12 slice 2)”; “refactor(split): extract dependencies, snapshot, lsp and repl crates (S15, CW-6129) \[skip ci\]”; “fix: show non-ASCII paths as written in /diff and git mentions (\#7317)” (refs \#7317).
crates/dependencies · medium confidence · unverified
36 commits (23 fixes) fixing crates/engine/src/fleet
A fix in crates/engine/src/fleet — 36 commits (23 fixs), 24 files: “fix: unblock contributor PR Rust validation \[skip ci\]”; “fix(engine,app-server): find Windows agent shims and keep probe refusals truthful (CW-7879)”; “fix(workspace): keep non-home link confinement, exempt only home state”; and 33 more (refs \#7293, \#7272).
crates/engine/src/fleet · medium confidence · unverified
49 commits (26 fixes) fixing crates/engine/src/core
A fix in crates/engine/src/core — 49 commits (26 fixs), 10 files: “fix: unblock contributor PR Rust validation \[skip ci\]”; “feat(engine): ask the user with options when a choice is theirs”; “fix(engine): bind stall supervision to admitted turn and offload blocking work”; and 46 more (refs \#6184, \#6553, \#7272).
crates/engine/src/core · medium confidence · unverified
5 commits (3 fixes) fixing crates/engine/src/rlm
A fix in crates/engine/src/rlm — 5 commits (3 fixs), 5 files: “test(engine,tui,plugins): align fixtures with the settled-usage accounting model”; “fix(engine): keep non-ASCII letters in derived RLM session names (\#7290)”; “fix(engine): retain captured usage and settle interrupted workers”; and 2 more (refs \#7290).
crates/engine/src/rlm · medium confidence · unverified
50 commits (26 fixes) fixing crates/engine/src/core/engine
A fix in crates/engine/src/core/engine — 50 commits (26 fixs), 23 files: “feat(acp): attach the extension host and configured MCP servers to ACP sessions (CW-7941)”; “feat(engine): ask the user with options when a choice is theirs”; “fix(engine): bind stall supervision to admitted turn and offload blocking work”; and 47 more (refs \#6184, \#6553, \#7272, \#7261).
crates/engine/src/core/engine · medium confidence · unverified
Decouple build version stamping from local git state and introduce release-only provenance SHA
The build-support library now prevents local, unstamped checkouts from triggering full workspace rebuilds by removing \.git\ file watchers and relying exclusively on explicit environment variables (\CODEWHALE\_BUILD\_SHA\, \DEEPSEEK\_BUILD\_SHA\, \GITHUB\_SHA\) for version stamps. Unstamped builds now display \(dev)\ instead of invalidating the cache, while packaged sources show the plain version. Additionally, a new \CODEWHALE\_RELEASE\_BUILD\_SHA\ is emitted only when explicit release variables are present, providing deterministic provenance for published binaries without network calls or local fallbacks.
crates/build-support/src · high confidence
Fix Windows CLI startup stack size and build rerun conditions
The CLI build script now configures the Windows main thread stack size to prevent startup failures caused by the default 1 MiB limit, and ensures the build reruns when the Git HEAD moves or the manifest directory changes, improving reliability in shared target directories.
crates/cli · high confidence
Fixes legacy config migration for top-level base\_url and api\_key
Resolves an issue where top-level \base\_url\ and \api\_key\ settings in legacy configuration files were not being correctly migrated to the new provider-specific structure. The fix ensures these values are properly moved into the \\[providers.deepseek\]\ table (or the appropriate vendor table based on the URL) during the configuration upgrade process, preventing misconfiguration after migration.
_crates/config/src/legacy\root · high confidence
Introduce strict path validation for Codewhale runtime directories
The new \crates/paths\ crate enforces that environment variables like \CODEWHALE\_HOME\ and \HOME\ resolve to absolute paths, rejecting relative values that could inadvertently place configuration or state files inside a repository. It also adds support for expanding leading \\~\ in path overrides and ensures the user home directory is resolved deterministically across platforms, preventing accidental leakage of sensitive data into local working directories.
crates/paths · high confidence
Sanitize and bound MCP server initialization instructions
The MCP client now sanitizes the \instructions\ provided by servers during initialization to ensure they are safe and bounded. Non-string values are ignored, control characters (except newlines and tabs) and bidirectional override characters are stripped, and the text is trimmed. If the resulting guidance exceeds 4096 bytes, it is truncated at a UTF-8 character boundary and appended with a \\[truncated\]\ marker, preventing potential injection or memory issues from untrusted server input.
crates/mcp · high confidence
Test coverage
1 commit adding/updating tests in crates/engine/src/client/chat, crates/engine/src/voice; 1 commit adding/updating tests in crates/skills/tests; 1 commit adding/updating tests in crates/tui-kit/tests; 1 commit adding/updating tests in crates/tui/tests/fixtures/conformance/responses; 15 commits adding/updating tests in crates/tui/src/tests; 18 commits adding/updating tests in crates/tui/tests; 2 commits adding/updating tests in crates/engine/src/task\_manager; 2 commits adding/updating tests in crates/tui/tests/fixtures; 3 commits adding/updating tests in crates/engine/src/automation\_manager; 3 commits adding/updating tests in crates/engine/tests; 4 commits adding/updating tests in crates/build-support/tests; 40 commits adding/updating tests in crates/engine/src/core/engine/tests; 41 commits adding/updating tests in crates/tui/tests/fixtures/conformance/events; 5 commits adding/updating tests in crates/app-server/tests; 50 commits adding/updating tests in crates/tui/src/tui/ui/tests; 7 commits adding/updating tests in crates/engine/src/pricing; 7 commits adding/updating tests in crates/engine/src/session\_manager; 8 commits adding/updating tests in crates/engine/tests/fixtures; 9 commits adding/updating tests in crates/engine/src/runtime\_threads/tests; Added Android test provider for live telemetry fixtures; Added PTY-based acceptance tests for TUI features; Added SSE conformance fixtures for Anthropic, DeepSeek, and OpenAI routes; Added acceptance tests for TUI core commands, session workflows, and tool lifecycle; Added comprehensive tests for transcript view caching and folding behavior; Added end-to-end tests for the Workflow JS runtime; Added parity tests for state store schema and thread operations; Added portable config group integration test; Added protocol parity tests for serialization and runtime envelopes; Added test support infrastructure for TUI integration testing; Added test to enforce single turn-loop architecture; Added tests for TUI doctor command security and configuration; Added tests for TUI frame composition and one-owner fact painting; Added tests for custom model configuration and legacy config compatibility; Added tests for exact Fleet schema and Reasoning Router integration; Added tests for mouse selection and primary clipboard behavior; Added tests for session group command handlers; Added tests for the Tideline footer and posture bar rendering; Added tests for the Tideline notifications inbox; Agent roster tests for usage, duration, and activity rendering; Expanded CLI integration tests for dispatch, diagnostics, and configuration; Expanded MCP conformance test suite; Expanded integration test coverage for TUI execution, diagnostics, and evaluation; Golden-buffer tests for the info line layout and content; New PTY-based integration test harness for TUI; Removed palette audit tests.
Dependencies
Patch crossterm 0.29.0 and unicode-width 0.2.2
The project now patches the crossterm terminal library to version 0.29.0 and the unicode-width crate to version 0.2.2. The crossterm update brings support for copying to the host clipboard via OSC52, improved keyboard enhancement flags, and updated dependencies like rustix 1.0 and mio 1.0. The unicode-width update ensures accurate column width calculations for Unicode characters, which is essential for correct text rendering in the terminal UI.
patches · high confidence
Workspace restructure and dependency consolidation for Codewhale v0.10.2
The project has been reorganized into a modular workspace of dedicated crates (e.g., \codewhale-agent\, \codewhale-cli\, \codewhale-runtime\, \codewhale-tui\) with internal dependencies pinned to version 0.10.2. This change consolidates the dependency graph, introducing specific library updates such as \rusqlite\ 0.40.2, \lru\ 0.18, \ratatui\ 0.30.2, and \rquickjs\ 0.12.0/0.14.0, while also adding the \@codewhale/extension-host\ npm package to support the TypeScript extension host.
(dependencies) · high confidence
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
How this codebase got here
Score
- CAI 57 → 61 (+4.1)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.4) — scores are not directly comparable.
Lenses
- Code Health 58 → 60 (+2.2)
- Architecture 83 → 81 (-1.5)
- Maturity 85 → 92 (+7.1)
- Readiness 59 → 62 (+3.0)
- Security 80 → 77 (-3.6)
- Domain Modelling 88 → 93 (+4.2)
- Event Sourcing 100 → 100 (+0.0)
- Accessibility 49 → 54 (+5.8)
- Performance 74 → 91 (+17.0)
Resolved (2672)
- (anonymous) (cognitive 36) (crates/tui/plugins/computer-use/app/daemon.mjs)
- (anonymous) (cyclomatic 21) (crates/tui/plugins/computer-use/app/daemon.mjs)
- AcpServer::project_event (cognitive 26) (crates/tui/src/acp_server/runtime.rs)
- AcpServer::project_event (cyclomatic 23) (crates/tui/src/acp_server/runtime.rs)
- AcpServer::project_turn (cognitive 63) (crates/tui/src/acp_server/runtime.rs)
- AcpServer::project_turn (cyclomatic 30) (crates/tui/src/acp_server/runtime.rs)
- AcpServer::replay_history (cognitive 65) (crates/tui/src/acp_server.rs)
- AcpServer::replay_history (cyclomatic 24) (crates/tui/src/acp_server.rs)
- AcpServer::serve (cognitive 37) (crates/tui/src/acp_server.rs)
- AcpServer::serve (cyclomatic 18) (crates/tui/src/acp_server.rs)
- AgentTool::execute (cognitive 18) (crates/tui/src/tools/subagent/mod.rs)
- AgentTool::execute (cyclomatic 20) (crates/tui/src/tools/subagent/mod.rs)
- AgentsCoordinateTool::mutate_coordination (cognitive 22) (crates/tui/src/tools/subagent/coord.rs)
- AgentsCoordinateTool::mutate_coordination (cyclomatic 20) (crates/tui/src/tools/subagent/coord.rs)
- AnchoredDirectory::open (cognitive 19) (crates/tui/src/tools/github/report.rs)
- ApprovalReplay::from_receipts (cognitive 18) (crates/tui/src/approval_log.rs)
- AutomationManager::get_runs_by_ids (cognitive 17) (crates/tui/src/automation_manager.rs)
- AutomationManager::list_runs_with_visibility (cognitive 17) (crates/tui/src/automation_manager.rs)
- AutomationManager::update_automation_unlocked (cognitive 26) (crates/tui/src/automation_manager.rs)
- AutomationManager::update_automation_unlocked (cyclomatic 23) (crates/tui/src/automation_manager.rs)
- …and 2652 more
New (3741)
- AccountClient::connect_models_with_cancel (cognitive 32) (crates/account-cli/src/session.rs)
- AccountClient::connect_models_with_cancel (cyclomatic 25) (crates/account-cli/src/session.rs)
- AccountLogin::poll_once (cognitive 17) (crates/account-cli/src/session.rs)
- AccountLoginView::handle_key (cognitive 18) (crates/tui/src/tui/account_login.rs)
- AccountLoginView::handle_key (cyclomatic 18) (crates/tui/src/tui/account_login.rs)
- AccountLoginView::render (cognitive 34) (crates/tui/src/tui/account_login.rs)
- AccountLoginView::render (cyclomatic 23) (crates/tui/src/tui/account_login.rs)
- AccountLoginView::tick (cyclomatic 20) (crates/tui/src/tui/account_login.rs)
- AcpClient::request (cognitive 24) (crates/engine/src/fleet/acp.rs)
- AcpServer::history_updates (cognitive 62) (crates/app-server/src/acp_server.rs)
- AcpServer::history_updates (cyclomatic 23) (crates/app-server/src/acp_server.rs)
- AgentTool::execute (cognitive 18) (crates/engine/src/tools/subagent/mod.rs)
- AgentTool::execute (cyclomatic 20) (crates/engine/src/tools/subagent/mod.rs)
- AgentsCoordinateTool::mutate_coordination (cognitive 22) (crates/engine/src/tools/subagent/coord.rs)
- AgentsCoordinateTool::mutate_coordination (cyclomatic 20) (crates/engine/src/tools/subagent/coord.rs)
- AnchoredDirectory::open (cognitive 23) (crates/engine/src/tools/github/report.rs)
- AnchoredDirectory::open (cyclomatic 17) (crates/engine/src/tools/github/report.rs)
- ApprovalReplay::from_receipts (cognitive 18) (crates/engine/src/approval_log.rs)
- AttachmentPreviewWidget::render (cognitive 31) (crates/tui/src/tui/widgets/attachment_preview.rs)
- AttachmentPreviewWidget::render (cyclomatic 18) (crates/tui/src/tui/widgets/attachment_preview.rs)
- …and 3721 more
Changes since last survey
- 300 commits — 189 feature/other, 111 fixes
By area
- crates/engine — 64 commits
- (root) — 48 commits
- web/lib — 32 commits
- crates/tui — 28 commits
- crates/app-server — 24 commits
- crates/plugins — 12 commits
- (repo) — 10 commits
- .github/workflows — 5 commits
- crates/localization — 5 commits
- web/data — 5 commits
- crates/config — 4 commits
- crates/sanitize — 4 commits
- crates/skills — 4 commits
- docs/RELEASE_CHECKLIST.md — 4 commits
- integrations/bridge-core — 4 commits
- scripts/release — 4 commits
- crates/runtime — 3 commits
- docs/zh_hans — 3 commits
- web/app — 3 commits
- crates/workflow — 2 commits
Notable commits
- fix: chore(changelog): sync the TUI changelog slice for the catalog, seam context and offline fix landing (CW-8038, CW-8012, CW-8031)
- fix: fix(acp): a child engine gets no MCP tools from the ACP tool build (CW-7941)
- fix: fix(acp): a plugin tool's nested approval reaches the editor as a permission request that names the nested tool (CW-8050)
- fix: fix(acp): a transport failure mid-turn still settles the turn and saves its checkpoint (CW-8043 repair 1)
- fix: fix(acp): refuse unsupported prompt blocks, relative cwd; classify write/edit tools; add stdio-style journey test (CW-7941)
- fix: fix(api): keep repeated hook labels distinct in plugin index
- fix: fix(app-server): discard a dead socket owner's record only when the store this start selected is gone (CW-8047)
- fix: fix(app-server): match a gone store spelled through a symlink or another letter case when retiring a dead owner receipt (CW-8054)
- fix: fix(app-server): plugin index names plugins by id, labels tools, signals cut name lists (CW-8049 PL-02 repair 1)
- fix: fix(app-server): read the original path of worktree-side renames in git status (#7310)
- fix: fix(app-server,runtime): quota route joins the test env scope; cap agent quota entries (CW-7993 AGENT-QUOTA)
- fix: fix(bridges): a chat keeps a migrated thread only when the bridge started it
- fix: fix(bridges): a chat that lost its thread at upgrade can start over, and recovery stops at unowned threads
- fix: fix(bridges): chats list and resume only the threads they created
- fix: fix(bridges): preserve pending-ownership recovery until Runtime confirmation settles
- fix: fix(catalog): Gemini 4 output limit is 65,536, sources cite the vendors (CW-8038)
- fix: fix(catalog): both model commands, the Runtime API, routing and cost read one captured catalog the same way
- fix: fix(ci): drop unused engine arboard and keep sccache off Windows legs
- fix: fix(ci): guard npm wrapper smoke sccache with matrix.os
- fix: fix(config): a named managed file that cannot be read is a load error (CW-8052)
- …and 280 more
Architecture
- Unchanged — 0 containers · 3 contexts · 0 edges
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
Survey your own repository
Hmbown/CodeWhale was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.
About this page
- The score is its most recent published measurement, taken on 14 November 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
- Measured at commit 059529678f5702ccf2c1ebd6c3ae52b3564e33f2 — the exact code this score is about.
- Scored under rubric-2026.11.4 — the same rubric and the same method as every other entry in this index.
- Measured by Canine Code Analyzer (cca.canine.dev) using cca measure-2026.11.65+preprod-36d4cd83012f.