v0.4.8

CCCC v0.4.8 Release Notes

v0.4.8 makes Web Pet materially more useful, tightens task and messaging behavior across Web and MCP, improves actor/runtime visibility, hardens projected browser flows, and adds another round of platform polish.

Highlights

1) Web Pet became a real attention assistant

Review scheduling, reminder generation, decision handling, and the Web Pet panel/bubble flow were substantially reworked so the system can surface clearer reminders, better suggestions, and more stable task proposals from live group state.

2) Task workflow handling is more reliable

Task handling across Web and MCP is more consistent, peer-created MCP tasks now default to self-assignment, and status/update flows are less fragile in day-to-day use.

3) Actor and runtime state are easier to trust

The actor list now exposes richer terminal-derived working state, startup paths gained stronger runtime preflight checks, and daemon transport diagnostics are surfaced more clearly for operators.

4) Browser-backed and auth flows were hardened

Projected browser support took a meaningful step forward in this release.

Browser-backed sessions used for embedded views and NotebookLM / Google auth are more robust, with better reconnect and compatibility behavior across the supported desktop environments.

5) Messaging, search, and Web behavior were cleaned up further

This release also includes a broad reliability pass across the Web and daemon surfaces.

v0.4.7

CCCC v0.4.7 Release Notes

v0.4.7 makes Presentation a first-class shared workspace, tightens task workflow handling across Web and MCP, adds Web branding controls, and cleans up several runtime and operator-facing rough edges.

Highlights

1) Presentation became a real shared workspace

The main change in v0.4.7 is the new Presentation surface.

Groups can now publish and manage slot-based presentation content through the daemon, Web UI, and MCP layer. The Chat surface also gained a dedicated Presentation rail and viewer flow, so shared artifacts no longer have to live only as chat text.

2) Interactive browser-backed views landed

v0.4.7 also adds the browser-surface path for Presentation.

This makes it possible to handle harder Web content in a more useful way than a simple static preview. The viewer lifecycle and controls were refined as part of the same work, including better handling for refresh, fullscreen, replacement, and URL entry.

3) Chat and Presentation are now connected by references

Presentation references were added to the message flow, which means chat can point back to a specific Presentation view instead of relying on vague text-only handoffs.

Snapshot and compare behavior were also added so the quoted view can be preserved more clearly when users and agents are discussing live content.

4) Task workflow handling is stricter

This release also improves the task workflow layer.

Task state handling in the Web UI is more structured, context/task workflow logic is tighter, and MCP-side task update compatibility was improved so task status changes are less fragile in practice.

5) Web polish and branding moved forward

v0.4.7 includes a broad Web polish pass as well.

Highlights include:

• Web branding controls for product name and logo assets
• stronger group/context/unread refresh behavior
• general message, panel, and console usability cleanup

6) Runtime and startup behavior were cleaned up further

This release also includes smaller but important runtime fixes.

Notably, the default cccc entry path now respects top-level --host / --port overrides throughout supervised Web startup and restart, and Kimi runtime defaults were updated to match the current preferred path.

v0.4.6

CCCC v0.4.6 Release Notes

v0.4.6 builds on v0.4.5 with a broader set of improvements than a pure prompt/preset polish release.

Compared with v0.4.5, this version extends IM bridge coverage, tightens Web-side context and cache behavior, introduces the first built-in role preset workflow, and hardens startup/shutdown behavior further across Windows and the supervised Web runtime.

Highlights

1) WeCom IM bridge support is now real and documented

The biggest functional addition in v0.4.6 is WeCom bridge support.

Key changes:

• a dedicated WeCom adapter was added instead of trying to force everything through the older bridge path
• Web-side IM bridge settings were extended so WeCom can be configured from the same operator surface as other bridge providers
• authentication and readiness behavior for the bridge were repaired and covered by targeted tests
• the docs now include a dedicated WeCom guide and screenshots for the operator flow

This matters because v0.4.5 still treated IM bridge support as narrower and more uneven in practice. v0.4.6 makes WeCom a real supported path instead of a partial edge lane.

2) Web context and cache behavior are tighter and less wasteful

v0.4.6 also improves how the Web surface reads and refreshes shared control-plane state.

Key improvements:

• actor and context route caching is more deliberate and more thoroughly tested
• cache invalidation after writes is less fragile, which reduces stale readback after actor/context updates
• related API/client/store coverage was expanded significantly so the Web state layer is less likely to drift silently
• surrounding settings and control surfaces were cleaned up so context-related interactions behave more predictably

This is not a flashy feature, but it is a meaningful quality improvement over v0.4.5 because it reduces a class of “the write succeeded but the UI still feels stale or noisy” problems.

3) Built-in role presets and prompt discipline landed

This release introduces the first built-in role preset workflow in the Web actor surfaces and strengthens the default collaboration guidance around those roles.

Highlights:

• built-in role presets were added for the first-wave roster, including planner, implementer, reviewer, debugger, explorer, and related roles
• preset application now gives actors a faster way to start from a strong role note instead of writing everything from scratch
• the prompt/help surface was tightened so startup guidance stays lean while richer discipline lives in the longer-lived help/preset layers
• planning, implementation, review, and debugging guidance are now sharper about things like existing-code leverage, full-diff-first review, regression thinking, and one-pass blast-radius cleanup

Relative to v0.4.5, this is the biggest day-to-day agent-behavior improvement in the release.

4) Windows and runtime lifecycle handling were hardened again

v0.4.5 already improved startup and shutdown behavior. v0.4.6 continues that line with another pass of runtime hardening.

Important changes:

• Windows shutdown cleanup was tightened further so lingering process/lifecycle edge cases are handled more reliably
• daemon/Web supervised runtime ownership and cleanup paths were hardened with additional lifecycle coverage
• Web readiness checks now tolerate more real failure modes, including OSError and HTTPException, instead of surfacing brittle behavior
• process-path handling and related platform-sensitive helpers received more validation, especially around mocked Windows path resolution and runtime probing

This makes v0.4.6 a stronger operational release than v0.4.5, especially if you are running CCCC on Windows or frequently exercising stop/start/restart flows.

5) Quality bar and verification

Beyond the visible changes above, v0.4.6 also includes:

• additional tests around Web cache behavior and stores
• stronger coverage for IM bridge startup/config flows
• targeted prompt-default and role-preset tests so the new guidance surfaces are less likely to regress silently
• follow-up lint/typecheck cleanup after the role preset work landed

In short, v0.4.6 is not just “the preset release.” It is the release that combines:

1. WeCom bridge support
2. better Web context/cache behavior
3. the first built-in role preset workflow
4. another round of Windows/runtime hardening

relative to the v0.4.5 baseline.

v0.4.5

CCCC v0.4.5 Release Notes

v0.4.5 consolidates more of the everyday workflow into the Web UI, simplifies remote access handling, and hardens runtime startup and shutdown behavior across Windows and POSIX environments.

Highlights

1) Web Pet replaced the old desktop pet surface

The old standalone desktop pet implementation was removed and its useful behavior was moved into the Web UI.

Key improvements:

• the pet now lives directly inside the Web interface instead of requiring a separate desktop/Tauri surface
• reminder bubbles and panel actions are more interactive, including direct jump targets such as opening chat or the relevant task
• the pet panel now shows task progress and smarter hints instead of only generic reminders
• agent tabs can surface the last terminal output after an agent stops, which makes “what just happened?” easier to answer without reopening logs

2) Web access and reachability are clearer

v0.4.5 continues the Web-access cleanup started in the previous line and makes the remote-access story more coherent in practice.

Highlights:

• the Web Access panel is better aligned with real operator goals such as local-only access, LAN/private access, and externally exposed access
• supervised Web restart/apply flows are clearer from the main cccc session
• the Web health endpoint is now publicly reachable, which makes health checks and external probing simpler
• surrounding UI wording and styling were refined so the current access posture is easier to understand at a glance

3) Runtime lifecycle behavior was hardened on Windows and POSIX

This release includes several fixes in the startup and shutdown path that materially improve day-to-day reliability.

Important changes:

• Windows Codex MCP setup is more robust and now prefers a stable absolute cccc entrypoint instead of trusting weaker path assumptions
• stale or misleading Codex MCP entries are less likely to be treated as “already installed”
• supervised Web child shutdown handling was tightened so Ctrl+C and restart flows behave more predictably
• POSIX background Python startup now preserves the active virtualenv interpreter path instead of resolving it to the system Python and silently losing the environment

4) The runtime support surface is less fragile

Beyond startup mechanics, v0.4.5 reduces a few sharp edges in the runtime control plane.

This includes:

• fail-fast MCP startup checks are scoped more carefully so they do not over-block unrelated lifecycle flows
• IM bridge child-process startup paths now follow the same safer background-process rules as the daemon/Web stack
• cross-platform process handling was refactored and validated more thoroughly in tests

v0.4.4

CCCC v0.4.4 Release Notes

v0.4.4 is a small polish release over v0.4.3.
It focuses on making the settings experience clearer, reducing UI/config friction, and aligning safer defaults with actual operator intent.

Highlights

• Guidance is now the default first-open tab in group settings, matching the visible tab order instead of dropping users into Automation first.
• Settings terminology now more clearly separates built-in automation from user-authored rules and snippets.
• The Delivery panel was simplified around the only user-facing behavior that still matters there: whether successful PTY delivery should auto-advance the read cursor.
• The Web settings UI no longer exposes min_interval_seconds, while daemon/API compatibility for that low-level delivery throttle remains intact.
• Actor idle alerts now default to 0 (off) for new/default/reset paths, without silently changing existing group configs that already store a value.

v0.4.3

CCCC v0.4.3 Release Notes

v0.4.3 is a major refinement release over v0.4.2.
It tightens the core collaboration model, reduces prompt and control-plane drift, hardens Windows and runtime integration, and makes long-running group operation more predictable across Web, MCP, IM, and Group Space workflows.

Highlights

1) Prompt, help, and role guidance were re-layered

The guidance stack was simplified so stable startup rules, live runtime guidance, and actor-specific notes no longer compete with each other.

Key changes:

• the startup prompt is slimmer and more focused on durable operating rules
• live runtime guidance now belongs to cccc_help, so optional capability-specific instructions appear only when actually relevant
• actor role notes are now canonically stored in the group help document @actor blocks instead of leaking into working-state fields
• agent working stance and role guidance are now more clearly separated, which reduces confusion between long-term role intent and current execution context

2) Context and task authority semantics were hardened

Several core collaboration boundaries were tightened so the system behaves more predictably under real multi-actor usage.

This includes:

• task.restore now follows a clear archived-only precondition and matches the same authority model as the other task-mutating operations
• peers can no longer mutate arbitrary unassigned tasks; task control is limited to the actor actually responsible for the work
• actor-owned agent_state semantics are now aligned across docs, daemon behavior, MCP tooling, and Web expectations
• system-driven task-to-working-state synchronization remains in place where it materially reduces manual upkeep cost

3) Group Space and memory workflows became more coherent

v0.4.3 significantly improves the operational behavior of notebook-backed workflows.

Key improvements:

• work-lane and memory-lane status now track the current binding more accurately instead of leaking stale sync residue after unbind/rebind cycles
• NotebookLM-related runtime guidance is injected only when the relevant capability is actually active
• the login, bind, use, unbind, and disconnect flow is cleaner from both the daemon and Web state-model perspective
• memory and daily-memory coordination stays intentionally connected, but the surrounding guidance and runtime status are now clearer

4) Runtime support and Windows robustness were strengthened

This release continues the push toward a more disciplined and supportable runtime surface.

Highlights:

• Windows MCP reliability was hardened, including runtime-context resolution and stdio/encoding robustness
• the officially supported runtime surface was narrowed and clarified to the runtimes CCCC can set up and operate reliably
• standalone Web startup now follows the same local-first binding model as the main CLI entrypoint
• release verification was tightened to better match the project's claimed Linux, macOS, and Windows support surface

5) Operator-facing surfaces saw broad stabilization

A large amount of polish landed across the everyday control surfaces used by operators and agent teams.

Notable areas:

• user-scoped actor profiles now work end to end across daemon, Web, and MCP paths
• IM integration continued to improve, especially around DingTalk sender identity, revoke behavior, @ targeting, and streaming fallback behavior
• blueprint export/import portability was fixed so round-trips keep the intended portable fields
• Web settings, modal behavior, overflow handling, context presentation, and translation coverage were substantially cleaned up
• global browser surfaces were trimmed so scoped users see less irrelevant machine-global data by default

Validation Summary

The 0.4.3 line was stabilized through multiple release candidates and targeted regression work across the highest-risk areas, including:

• context and task authority rules
• role-notes source-of-truth handling
• Group Space bind/unbind status correctness
• Windows MCP stability
• scoped actor profiles and permission boundaries
• IM bridge revoke and delivery behavior
• blueprint round-trip portability

Upgrade Notes

From v0.4.2, this is still a drop-in minor upgrade for most deployments.
No explicit data migration step is required.

Recommended post-upgrade checks:

1. restart the daemon and Web once to ensure all runtime surfaces are on the new prompt/help/control-plane behavior
2. if you use custom help or prompt overrides, re-check your local overrides against the new guidance layering
3. if you use Group Space or NotebookLM workflows, verify current lane bindings and status after the upgrade
4. if you use scoped users, actor profiles, or remote/browser access, verify one non-admin path end to end

desktop-v0.0.3

Full Changelog: desktop-v0.0.2...desktop-v0.0.3

desktop-v0.0.2

Full Changelog: desktop-v0.0.1...desktop-v0.0.2

desktop-v0.0.1

What's Changed

• fix(web): reload events on SSE reconnect to fill disconnect gap by @howold-lab in #47
• fix(im/discord): resolve three issues preventing Discord bridge from working by @KQAR in #48
• Pr55 history integration by @ChesterRa in #56
• windows适配支持 by @kinghy949 in #55
• Topic/current context prompt pass by @ChesterRa in #57
• fix(encoding):convert files from UTF-8 to UTF-8 BOM. by @yuansui486 in #58

New Contributors

• @KQAR made their first contribution in #48
• @ChesterRa made their first contribution in #56
• @kinghy949 made their first contribution in #55

Full Changelog: v0.4.2...desktop-v0.0.1

v0.4.2

CCCC v0.4.2 Release Notes

v0.4.2 advances the IM integration and operator-facing control surfaces.
It ships key-based IM chat authorization, a reusable actor profiles system, and a remote access control plane — all with matching Web/MCP/CLI coverage.

Highlights

1) Key-based IM chat authorization

IM bridges now authenticate chats through a cryptographic bind-key flow instead of static trust.
Operators generate a one-time bind key in the Web UI, share it with the target IM chat, and the bridge verifies identity before granting access.

Key capabilities:

• /bind <key> command in supported IM platforms (Telegram first)
• automatic subscription on successful bind (no separate /subscribe step)
• pending approval management: operators can review and approve/deny bind requests from the Web Settings → IM Bridge tab
• revoke semantics: authorized chats can be revoked at any time, with immediate effect on message routing
• cccc_im_bind MCP tool for programmatic binding from automation workflows

2) Actor profiles

A new actor profile abstraction lets operators attach reusable identity metadata (display name, avatar, prompt preamble, model preferences) to actors.

Profiles are linked at actor creation or update and propagate across:

• daemon runtime (used for display in IM bridges and Web)
• Web UI (dedicated Actor Profiles settings tab for CRUD operations)
• MCP surface (profile query and assignment tools)

3) Remote access control plane

A new remote access layer allows authorized clients to interact with a CCCC daemon across network boundaries.
Combined with hardened IM revoke semantics, this enables secure multi-node orchestration without exposing the daemon socket directly.

4) Telegram UX improvements

• Typing indicators: the bridge now sends typing actions with configurable throttling, giving collaborators visual feedback during agent processing.
• Actor titles over IDs: IM-rendered messages now prefer human-readable actor titles instead of raw actor identifiers.

5) MCP and operational fixes

• Recipient semantics fix: None (broadcast) vs [] (no recipients) distinction is now correctly preserved in MCP message send, preventing unintended broadcast or silent drops.
• IM KeyManager reload: authorization state is reloaded from disk on each inbound poll, eliminating stale-cache authorization failures after daemon restarts.
• Timestamp handling: authorized_at timestamps in the IM Web UI are now correctly parsed and displayed.