CHANGELOG.md

Changelog

All notable changes to this project are documented in this file.

The format follows Keep a Changelog, and versions follow SemVer/PEP 440.

[0.4.12] — 2026-04-23

Added

• Voice Secretary workspace with repository-backed markdown documents, Document/Ask/Prompt modes, request history, transcript feedback, document archive/download actions, and dedicated assistant reporting paths.
• Built-in assistant controls in the chat composer for PET and Voice Secretary, plus a dedicated assistant settings surface with larger prompt editors.
• Daemon-owned tracked delegation with task/message linkage, idempotency hardening, and task chip projection in chat messages.
• cccc update command with install-source detection.

Changed

• Voice Secretary input handling now routes stable transcript/request data through dedicated assistant surfaces instead of noisy chat-style JSON notifications.
• Headless delivery and inbound rendering were tightened for Codex and Claude, including clearer sender/recipient context and safer reply routing.
• Runtime Dock state projection was polished so PTY/headless actors and built-in assistants expose clearer idle, active, and stopped states.
• Web workspace and settings UI received broad composer, modal, assistant, capability, automation, copy, and markdown rendering polish.
• Collaboration follow-up settings were renamed and regrouped around operator-facing behavior instead of internal automation terminology.

Fixed

• Fixed partial-failure retry handling for tracked delegation so retries do not duplicate tasks after message-send failures.
• Fixed Voice Secretary prompt-refine, Ask reply, idle-review, and document routing edge cases.
• Fixed headless reply recipient routing for ambiguous sender/source contexts.
• Fixed user message bubble background regression after text-color unification.
• Fixed actor secret placeholder examples for Codex and Claude Code so the UI no longer suggests ineffective OpenAI environment variables for Codex.
• Fixed WeCom response URL fallback behavior for streaming and media replies.

[0.4.11] — 2026-04-11

Added

• Weixin bridge now uses the Python wechatbot-sdk integration instead of the previous packaged Node.js sidecar, reducing bundled bridge assets and keeping Weixin login, inbound, media, and outbound behavior inside the Python adapter stack.
• Weixin subscription guidance in Web settings: after Weixin login, the IM Bridge panel now explicitly explains the /subscribe and Pending Requests flow, including separate guidance for unconfigured, stopped, already-bound, and ready-to-subscribe states.
• Runtime Dock ring tone coverage extracted into a dedicated helper so PTY and headless actors share clearer stopped, ready, queued, active, and attention state mapping.
• Mention suggestion labels in chat composer now show actor display labels with secondary IDs where useful, and the mention preview can be closed with Escape.

Changed

• Weixin packaging was simplified by removing the Node sidecar packages and packaged .mjs resources; wechatbot-sdk>=0.2.0 is now the Python dependency for the Weixin path.
• Headless streaming reconciliation was tightened so pending placeholders, canonical reply sessions, stream-id promotion, and terminal reply phases are less likely to reset or regress after final replies.
• Runtime state projection now writes stopped actor entries to the ledger when actors disappear from runtime snapshots, helping the Web clear stale working halos and live indicators.
• Actor edit modal synchronization was cleaned up so profile-backed actors and custom actors open with settings that better match their current stored configuration.
• Web group runtime updates now use SSE/runtime projections more consistently, reducing sidebar state drift after refreshes or lifecycle changes.

Fixed

• Fixed CLI daemon fallback behavior so daemon rejections are not incorrectly treated as permission to fall back to local mutations.
• Fixed Weixin outbound/context-token handling so cached SDK context can be rehydrated into the running bot and outbound readiness survives bridge restarts more reliably.
• Fixed Weixin IM configuration canonicalization for empty and legacy account fields, with additional route and adapter coverage.
• Fixed missing visibility for actor.activity ledger append failures by logging append errors instead of silently swallowing them.
• Fixed EventKind documentation parity gaps by adding internal contract coverage.
• Fixed a bare Chinese placeholder in the Weixin settings panel by moving it into the English, Chinese, and Japanese locale files.

[0.4.10] — 2026-04-10

Added

• Claude headless runtime support alongside the generalized headless streaming pipeline, enabling structured headless sessions beyond Codex.
• Runtime Dock and live trace surfaces in Web: richer headless previews, compact activity timelines, grouped runtime inspectors, and better runtime state projection across chat and actor views.
• Headless runtime plumbing and test coverage: cache/projection helpers, broader coverage for headless events, runtime startup, Web actor routes, and Windows PTY behavior.
• Weixin sidecar support refreshed with expanded IM bridge adapter-level validation.

Changed

• Headless delivery architecture generalized from Codex-specific to a shared headless model used consistently across daemon, Web, and MCP surfaces.
• Web chat and runtime UX significantly refined: message reconciliation, activity persistence, runtime previews, composer behavior, per-message identity rendering, and group runtime controls.
• Workspace and presentation flows streamlined through enhanced browser/presentation handling and more resilient scope attachment behavior.
• CI and release verification strengthened with longer timeout coverage, pytest-timeout in smoke paths, and tighter release workflow checks.

Fixed

• Fixed multiple headless reply and streaming lifecycle regressions, including fallback flow errors, message identity drift, canonical-reply reconciliation, and startup-state mismatches.
• Fixed missed headless injection for automation-generated system.notify events — automation-triggered notifications now reach running headless agents instead of only landing in the inbox.
• Fixed PTY teardown hardening, unread/index parity edge cases, and additional delivery flow stabilization.
• Fixed Web chat rendering regressions: lost avatars, unstable activity bubbles, stale runtime preview state, and composer alignment under text scaling.
• Fixed group start/pause button UI not updating — setGroupDoc now syncs runtime_status.lifecycle_state immediately, and refreshGroups patches runtime_status and running from server meta so stale local state no longer overrides the authoritative status.
• Fixed Windows PTY wake-path locking problems and related test instability.
• Fixed IM integration reliability: DingTalk mention handling, Weixin sidecar SDK pinning, and adapter behavior.

[0.4.9] — 2026-04-05

Added

• WeChat (Weixin) IM bridge: Node.js sidecar, CLI login/logout, QR-code auth in Web UI, and daemon routes for bridge lifecycle, following the same bind-key authorization model as other adapters.
• Text-size accessibility control: three-tier scale selector (90% / 100% / 125%) persisted per-browser. System-theme icon changed to a display icon; mobile menu now cycles light → dark → system.
• Async result contract: formal async_result IPC signaling (accepted/completed/queued) across daemon actor operations.
• Assistive-jobs layer: explicit pet review and profile job kinds requiring verified completion before marking done.

Changed

• Actor launch pipeline unified: add/update/lifecycle/runtime operations now share one resolution path with consistent async-result semantics; group start/stop reliably awaits per-actor results.
• Pet runtime tracks group settings: desktop-pet enablement syncs with group-settings changes.
• Web Pet task advisor is local-first: local evidence evaluation before surfacing proposals, reducing speculative noise.
• Presentation viewer gained inline web-preview support, topic-aware slide navigation, and split-layout mode for simultaneous conversation and viewing.
• Group sidebar extracted into a standalone component with optimized chunk splitting.
• MCP dynamic capability tools reflect real-time actor state.

Fixed

• Fixed idle-standup suppression and silence-activity filters to stay quiet when there is genuinely nothing to act on.
• Fixed runtime visibility controls so peer and pet tabs show/hide based on actor composition.
• Fixed context sync and group-space writeback to distinguish accepted vs. completed status.
• Fixed automation snippet catalog separation so built-in overrides are distinct from user-authored rules.

[0.4.8] — 2026-03-30

Added

• Windows-friendly env snippet support in the Web secret editor: set KEY=VALUE and $env:KEY="VALUE" forms accepted alongside Unix-style entries.
• Terminal-derived working state exposed in the actor list for richer runtime visibility.
• Modularized Web API services for a cleaner frontend integration layer.

Changed

• Web Pet substantially reworked: review scheduling, reminder generation, decision handling, and task proposals are more reliable and less noisy.
• Actor startup gained stronger runtime preflight checks and clearer daemon transport diagnostics.
• Peer-created MCP tasks now default to self-assignment instead of unassigned.
• Ledger and unread-index paths made faster with reduced overhead.

Fixed

• Fixed projected browser session reliability for embedded views and NotebookLM/Google auth flows.
• Fixed task status and update flows being fragile under rapid MCP task operations.
• Fixed Web-to-daemon messaging semantics and context/chat UI behavior after the Web API modularization.

[0.4.7] — 2026-03-23

Added

• Presentation workspace: slot-based presentation content managed through daemon, Web, and MCP, with a dedicated Chat Presentation rail and viewer flow.
• Browser-backed presentation views: interactive viewer lifecycle with refresh, fullscreen, replacement, and URL entry.
• Presentation references in chat: messages can point to a specific Presentation view with snapshot and compare support.
• Web branding controls: product name and logo asset configuration from the Web settings surface.

Changed

• Task state handling in Web UI is more structured; task/context workflow logic is tighter.
• MCP task update compatibility improved so status changes are less fragile.
• Default cccc entry path now respects top-level --host / --port overrides throughout supervised Web startup and restart.
• Kimi runtime defaults updated to match the current preferred path.

Fixed

• Fixed group/context/unread refresh behavior for better state coherence post-mutation.
• Fixed general message, panel, and console usability issues across the Web surface.

[0.4.6] — 2026-03-19

Added

• WeCom IM bridge: dedicated adapter, Web-side bridge settings, authentication/readiness behavior, and operator docs including a dedicated WeCom setup guide.
• Built-in role presets: first-wave roster (planner, implementer, reviewer, debugger, explorer) with a faster preset-application UI for common actor role starting points.

Changed

• Web context and actor route caching made more deliberate with proper invalidation after writes, reducing stale readback after actor/context updates.
• Prompt and help surface tightened so startup guidance stays lean; richer guidance lives in the help/preset layers.
• Web readiness checks now tolerate OSError and HTTPException instead of surfacing brittle failure behavior.

Fixed

• Fixed Windows shutdown cleanup for lingering process/lifecycle edge cases.
• Fixed cache invalidation after actor/context writes to prevent stale Web UI state.
• Fixed WeCom adapter startup and config flows.

[0.4.5] — 2026-03-18

Added

• Web Pet panel: task progress, smarter hints, direct jump to chat/task, and post-stop terminal output snippet after an agent ends a session.
• Web health endpoint made publicly reachable for external health checks and probing.
• Supervised Web restart/apply flow surfaced more clearly from the main cccc session.

Changed

• Desktop pet surface removed: Web Pet is now the primary pet surface (previous Tauri-based implementation retired).
• Web Access panel better aligned with real operator goals: local-only, LAN/private, and externally exposed access postures are clearer.
• POSIX background Python startup now preserves the active virtualenv interpreter path instead of resolving to system Python.

Fixed

• Fixed Windows Codex MCP setup to prefer a stable absolute cccc entrypoint and avoid false "already installed" detection.
• Fixed supervised Web child shutdown so Ctrl+C and restart flows behave predictably.
• Fixed fail-fast MCP startup checks to avoid over-blocking unrelated lifecycle flows.
• Fixed IM bridge child-process startup to follow the same safer background-process rules as the daemon/Web stack.

[0.4.4] — 2026-03-16

Changed

• Group settings: Guidance is now the default first-open tab, matching the visible tab order.
• Settings terminology now more clearly separates built-in automation from user-authored rules and snippets.
• Delivery panel simplified to the only user-facing behavior that remains: PTY delivery auto-advance of the read cursor.
• Actor idle alerts default to 0 (off) for new/default/reset paths without silently changing existing stored values.

Removed

• min_interval_seconds removed from the Web settings UI (daemon/API compatibility preserved).

[0.4.3] — 2026-03-15

Added

• User-scoped actor profiles working end-to-end across daemon, Web, and MCP paths.
• NotebookLM runtime guidance injected into the help layer only when the relevant capability is actually active.

Changed

• Guidance stack re-layered: startup prompt is slimmer; live capability guidance is in cccc_help; actor role notes are canonically stored in group help @actor blocks instead of leaking into working-state fields.
• Task authority model tightened: task.restore follows an archived-only precondition; peers can no longer mutate unassigned tasks outside their own scope.
• agent_state semantics aligned across docs, daemon behavior, MCP tooling, and Web expectations.
• Group Space bind/unbind status now tracks the current binding accurately without leaking stale sync residue after rebind cycles.
• Runtime support surface narrowed to runtimes CCCC can set up and operate reliably; standalone Web startup follows the same local-first binding model as the main CLI.

Fixed

• Fixed blueprint export/import round-trips so portable fields survive a full cycle without divergence.
• Fixed Windows MCP reliability: runtime-context resolution and stdio/encoding robustness.
• Fixed DingTalk sender identity, revoke behavior, @ targeting, and streaming fallback edge cases.
• Fixed Web settings, modal overflow, context presentation, and translation coverage gaps.
• Fixed global browser surfaces to default-scope users to relevant data, reducing machine-global noise for scoped users.

[0.4.2] — 2026-02-22

Added

• IM key-based chat authorization: dynamic bind-key authentication for IM bridges, replacing static trust with per-chat cryptographic binding. Includes /bind command, auto-subscribe on successful bind, pending approval management, and revoke semantics.
• cccc_im_bind MCP tool: programmatic chat authorization via the MCP surface.
• Authorized chats Web UI: view/manage bound IM chats and pending bind approvals from Settings → IM Bridge tab.
• Bind key UI: generate and display bind keys for chat authorization directly from Web settings.
• Actor profiles system: profile linking across daemon, Web, and MCP — including profile runtime, persistent store, and a dedicated Actor Profiles settings tab.
• Remote access control plane: remote daemon access with hardened IM revoke semantics for secure multi-node operation.
• Telegram typing indicators: typing action support with configurable throttling for more natural conversational UX.
• IM authentication IPC documentation: new standards doc covering IM auth IPC methods.

Changed

• IM display names: prefer actor titles over raw actor IDs in IM-rendered messages for better readability.
• Modal UX refinements: extracted modal close handlers and adjusted inbox modal height for cleaner interaction.

Fixed

• Fixed MCP message send incorrectly collapsing None recipients to empty list, breaking broadcast semantics.
• Fixed authorized_at timestamp handling in IM Web UI.
• Fixed IM KeyManager state not reloading from disk on each inbound poll, causing stale authorization data.
• Fixed docs incorrectly requiring post-bind /subscribe step (now handled automatically).

[0.4.1] — 2026-02-20

Added

• Actor lifecycle event coverage: daemon streaming now emits fuller actor/group lifecycle transitions for better observability and downstream automation hooks.
• Secret safety UX upgrade: actor secret keys now support masked previews in Web edit flows, improving operator confidence without exposing plaintext.
• Branding assets refresh: project logos were added and integrated into Web/README surfaces for consistent distribution identity.

Changed

• Automation idle semantics were aligned with explicit group-state behavior, reducing ambiguity in scheduled reminder execution during idle mode.
• Terminal safety hardening: resize and related terminal maintenance paths were tightened to avoid unstable behavior in mixed runtime conditions.
• Docs and onboarding were updated to match current v0.4 behavior, including SDK entry points, release hub linking, and refreshed top-level README content.

Fixed

• Fixed lifecycle edge cases where actor/group state transitions and event emission could diverge.
• Fixed multiple test-surface instability points (especially MCP/environment isolation), improving release reproducibility.

[0.4.0] — 2026-02-16

Added

• Chat-native orchestration model: operators can assign and coordinate work in a persistent Web conversation, with full delivery/read/ack/reply state tracking.
• External IM extension of the same workflow: Telegram, Slack, Discord, Feishu/Lark, and DingTalk bridges allow the same group control model outside the browser.
• Prompt-configurable multi-agent workflow design: guidance prompts and automation rules become first-class workflow controls instead of ad-hoc conventions.
• Bi-directional orchestration capability: CCCC schedules agents, and agents can schedule/manage CCCC workflows via MCP tools under explicit boundaries.
• Append-only ledger truth model: every group event is persisted in groups/<group_id>/ledger.jsonl for replayable, auditable operations.
• Structured automation engine: interval/recurring/one-time triggers with typed actions (notify, group_state, actor_control) for operational delegation.
• Accountable messaging semantics: read cursors, acknowledgement paths, and reply-required obligations for high-signal collaboration.

Changed

• Generation shift from v0.3: replaced the tmux-first operating model with a daemon-first collaboration kernel and versioned contracts.
• Control-plane unification: Web/CLI/MCP/IM now operate on one shared state model (thin ports, daemon-owned truth).
• Runtime state standardization: operational state is managed under CCCC_HOME (default ~/.cccc/) instead of repository-local state.
• Operating workflow modernization: day-to-day usage aligns around attach / actor / group / send / mcp over tmux-era command patterns.

Fixed

• Reliability hardening across RC cycles for delivery flow, automation execution, reconnect/resume handling, and registry normalization.
• Stability and UX fixes in Web interactions (including mobile operation and composer/tasking flows).
• MCP/docs/CLI parity drift reduced through dedicated guardrail tests.

Removed

• Deprecated tmux-first orchestration line from active mainline development (archived at cccc-tmux).

[0.4.0rc21] — 2025-07-24

Added

• Web i18n framework: integrated react-i18next with namespace-based locale loading (common, layout, chat, modals, actors, settings) and automatic browser language detection.
• Chinese (zh) locale: complete Simplified Chinese translation across all 6 namespaces (735 keys), with native-level phrasing review and unified typography (full-width ：, Unicode …).
• Japanese (ja) locale: complete Japanese translation across all 6 namespaces (735 keys), with native-level phrasing review and unified typography (full-width ：, Unicode …, full-width ？).
• Language switcher UI: minimal trigger button showing only short label (EN/中/日), positioned at the rightmost of the header; dropdown panel with scale-in animation and left accent bar for active item; React Portal for proper positioning.
• i18n key parity test: automated test to verify all locale files have identical key sets across languages.

Changed

• LanguageSwitcher refactored from cycle-button to professional popover dropdown.
• Language switcher moved to header rightmost position (after Settings button) with separator.
• Shared language configuration extracted to languages.ts.
• README overhaul with comprehensive project details, architecture, features, and quick start guide.
• Installation instructions updated to use TestPyPI for release candidates.
• Docker Claude config updated with bypass permissions flag.

Fixed

• Chinese locale encoding unified from \uXXXX escape sequences to direct Unicode characters.
• Chinese translation quality: 忙碌中→处理中, 义务状态→回复状态, 编辑器→输入框, 代码片段→模板.
• Japanese colon typography: 39 instances of half-width : after CJK characters corrected to full-width ：.
• to label in chat kept as English "To" in ZH/JA (international convention).
• Misleading "Clipboard" label corrected to "Context" in EN/ZH/JA layout.

[0.4.0rc20] — 2026-02-13

Added

• Daemon modularization: extracted monolithic server.py into 22+ focused ops modules with full dispatch orchestration (request_dispatch_ops.py), preserving identical logic with callback injection for all external dependencies.
• MCP parity guardrails: toolspec dispatch parity test, schema guard test, CLI reference parity test, and web automation docs parity test.
• MCP toolspec normalization: consistent indentation and formatting across all 1400+ lines of tool definitions.
• Web UI component extraction: ModalFrame, SettingsNavigation, ContextSectionJumpBar, ProjectSavedNotifyModal, ScopeTooltip extracted from monolithic modal files.
• Docker deployment guide with custom API endpoint configuration and proxy handling.
• Access-token authentication gate and non-root Docker user support.
• Auto-wake disabled recipients: agents are automatically started when they receive a message.
• Message mode selector in the Web UI composer; reply-required workflow with digest nudges.
• DingTalk enhancements: message deduplication, file sending via new API, stream mode documentation.
• IM bridge improvements: proxy environment variable passthrough, implicit send behavior clarification.
• cccc_group_set_state MCP tool now accepts stopped (mapped to group_stop).

Changed

• Daemon ops modules use dependency injection (callbacks) instead of global imports for testability.
• Serve loop extracted to serve_ops.py; socket protocol to socket_protocol_ops.py.
• MCP dispatcher split by namespace; tool schemas extracted from handler code.
• Web Settings: AutomationTab split into focused subcomponents.
• Runtime behavior tests made runner-independent for cross-platform CI.
• Registry auto-cleans orphaned entries on group load failures.
• Release and standards docs aligned to version-agnostic examples.

Fixed

• Orphaned PTY actor processes cleaned up on daemon restart.
• Mobile modal UX regressions (composer, runtime selectors).
• Template import now correctly handles auto_mark_on_delivery.
• Tooltip ref callback stability in Web UI.
• reply_required correctly coerced to boolean in MCP message send.

[0.4.0rc18]

Notes

• Release candidate baseline before the rc19/rc20 quality-convergence cycle.
• Established append-only ledger, N-actor model, MCP tool surface, Web UI console, and IM bridge architecture.