CCCC v0.4.27 Release Notes

v0.4.27 is a collaboration-surface cleanup and ChatGPT Web Model usability release.

Compared with v0.4.26, this release makes CCCC simpler to operate in three practical ways: the old PET/WebPet mechanism is gone, ChatGPT Web Model setup is easier to understand and less coupled to fragile ChatGPT page details, and agents can now suggest the user's next message without sending it automatically.

A Simpler Built-In Assistant Model

CCCC no longer ships PET or WebPet as active product surfaces.

This removes the PET daemon/runtime path, WebPet UI, PET-specific settings, MCP tool exposure, generated WebPet assets, local task proposal loops, review reminders, and PET context refresh machinery. Voice Secretary remains the supported built-in assistant path.

The cleanup is intentionally conservative around history and compatibility:

• historical release notes and changelog entries still mention PET where they document old behavior;
• legacy ## @pet help blocks are preserved as legacy-tagged content instead of becoming active prompts;
• stale internal actor records are treated as unsupported internal actors and skipped during group start/autostart.

For users, the result is a smaller and less ambiguous product: CCCC no longer has two overlapping built-in assistant concepts competing for attention.

Clearer ChatGPT Web Model Setup

The ChatGPT Web Model settings flow has been reorganized around the order users actually need to follow:

1. configure public Web Access;
2. create a ChatGPT Web Model actor in the target group;
3. sign in to ChatGPT in the embedded browser;
4. connect the CCCC MCP app in ChatGPT;
5. save the ChatGPT delivery target.

The Web Model settings panel now focuses on the ChatGPT-specific steps. Web Access remains in the Web Access settings tab, and actor creation/startup remains in the group actor controls, where the target group is explicit.

Target selection is also more direct. CCCC can now use a saved conversation URL, save the current inspected ChatGPT conversation, or start a new chat on the next delivery and bind it once ChatGPT creates the final chatgpt.com/c/... URL. Diagnostic browser history such as last_tab_url is no longer treated as a delivery target.

Less Fragile ChatGPT Browser Delivery

ChatGPT has changed its app permission UI, including support for persistent approval such as "Always allow" in supported accounts. CCCC now leans into that model instead of trying to chase ChatGPT page internals.

Automatic ChatGPT page refresh recovery is disabled by default. The mechanism is still available for fragile browser sessions by setting:

export CCCC_WEB_MODEL_BROWSER_AUTO_RELOAD=1

CCCC also no longer tries to click ChatGPT permission prompts. Users should approve the CCCC app in ChatGPT directly, preferably with "Always allow" when they trust the local connector. When a browser-delivered message appears to be waiting on ChatGPT-side app approval, CCCC Web shows a lightweight hint so the user knows to check the ChatGPT page instead of waiting silently.

New-chat delivery binding has been tightened as well. When "start new chat on next delivery" is selected, CCCC records the delivery, waits for ChatGPT to expose the concrete conversation URL, and persists that bound URL for future deliveries.

Suggested Next User Messages

Agents can now include an optional suggested_user_message when they send or reply to the user.

CCCC Web displays that suggestion inside the composer as gray editable prefill text. The user can:

• press Tab to accept it;
• click the inline suggestion button;
• edit the accepted text before sending;
• ignore it and type something else.

Suggestions are never sent automatically. They are designed to reduce repetitive manual typing while keeping the user in control.

The feature is available through daemon messaging, Web messaging, and MCP messaging APIs. It is intentionally scoped to user-facing messages: suggestions are hidden after the user has already replied, and they are not shown when the composer is routed to another group.

Smaller Fixes and Guardrails

This release also includes focused reliability and clarity fixes:

• ChatGPT target setup no longer defaults to saving stale browser history as the delivery target.
• The current ChatGPT target form is easier to read and uses the saved target as the starting point.
• Suggested messages are dismissed once the user accepts them, types over them, or dismisses them.
• Suggested-message freshness is based on the live group timeline rather than the currently filtered chat window.
• Cross-group composer routing suppresses suggestions so a proposal intended for one group cannot be sent into another.

Validation

The release was validated with focused backend and frontend coverage for PET removal boundaries, legacy help parsing, unsupported internal actors, ChatGPT target drafting, browser recovery behavior, app permission hints, new-chat binding, suggested-message contracts, and composer suggestion freshness/routing.

Why Upgrade

Upgrade to v0.4.27 if you want a cleaner CCCC without the old PET/WebPet layer, use ChatGPT Web Model and want a clearer setup/targeting flow, or want agents to provide editable next-message suggestions that speed up user replies without taking control away from the user.