# agnostic-ai v0.94.0: start, rename and delete chats from one place

> Source: <https://agnostic-ai.org/updates/2026-11-12-v0.94.0/>
> Published: 2026-11-11 23:00:00+00:00

## Highlights

- **Breaking:**`validate` fails a hook scoped with`targets:` to a tool that does not know its event.`sync` used to drop it silently. See[hook fields](https://agnostic-ai.org/docs/spec-format/hook-reference/#fields) .
- `sync` no longer loses a tool's rules because of`outputs.<target>.rules-file` , and stops with[AAI-102](https://agnostic-ai.org/docs/errors/#aai-102-targets-emit-to-the-same-output-path) where it would.
- Delete an archived Claude Code chat for good from the [companion](https://agnostic-ai.org/docs/cli-reference/companion/#delete-an-archived-chat) .
- [`agnostic-ai conversations start`](https://agnostic-ai.org/docs/cli-reference/conversations/#conversations-start) opens a linked chat in a new terminal tab, from inside Claude Code, Codex or`cursor-agent` .
- The companion and the web interface show the agent's latest reply or question under each chat's title.

## Upgrading from v0.93.0

- **Hooks scoped with `targets:`:** run` agnostic-ai validate` . For each hook it rejects, use that tool's own event name, or set the event under`x-<target>` , such as`x-augment: {event: PromptSubmit}` . See[hook fields](https://agnostic-ai.org/docs/spec-format/hook-reference/#fields) .
- **Teams that share a project:** upgrade every copy of agnostic-ai that syncs it.`sync` ,`doctor --fix` and`revert` now stop on sync state written by a newer release.

Then follow the [upgrade page](https://agnostic-ai.org/docs/cli-reference/upgrades/#upgrade).

## What changed

### Start a linked chat

Run `agnostic-ai conversations start` inside Claude Code, Codex or `cursor-agent` to open a new chat in a new tab of the same folder, with a short brief as its first prompt. The new chat shows indented under the one it came from, in both the web interface and the companion. Turn on the `side-conversation` built-in to give the agent a skill that says when to start one and how to write the brief. See [side conversations](https://agnostic-ai.org/docs/side-conversation/).

The new `wrap-up` built-in adds a skill that ends a session with what changed, what is still open, and a note to share. See [session wrap-ups](https://agnostic-ai.org/docs/wrap-up/).

### Keep track of each chat

- Each local Claude Code, Codex and Cursor chat shows the agent's question or last message under its title. Titles show what you typed, without the tags tools add. A Claude Code session with Remote Control on shows as one row, marked Shared.
- Rename a chat from its pencil button, with F2 or from the right-click menu, where you can also mark it as read. **Mark all as read** clears every unread or waiting chat.
- Hide takes the companion off the screen until the shortcut or a waiting chat brings it back. `upgrade` also updates an installed companion. See[upgrade the companion](https://agnostic-ai.org/docs/cli-reference/companion/#upgrade-the-companion) .

### Clean up finished chats

Delete an archived Claude Code chat from the companion's right-click menu. This removes its transcripts, tool output and backups. A chat that went quiet on its own can be deleted too, unless it was written to since you last looked. Other tools say why their chats cannot be deleted. See [delete an archived chat](https://agnostic-ai.org/docs/cli-reference/companion/#delete-an-archived-chat).

### Sync without losing files

- `sync` ignores`rules-file` on a tool that writes no merged rules file, and stops with AAI-102 when the setting would overwrite another tool's rules. See[entry-point files](https://agnostic-ai.org/docs/configuration/#entry-point-files) .
- `sync` and`status` also stop when an output sits on a backup file that sync saved, or when two outputs differ only in letter case, such as`agents.md` and`AGENTS.md` .
- `revert --force` no longer deletes a shared rules file it just restored.

See the [release notes](https://github.com/Chemaclass/agnostic-ai/releases/tag/v0.94.0) for the remaining fixes, including Augment hook payloads, Qwen hook timeouts and Copilot effort values.

## What changed in the targets

- **Augment:**`sync` writes hooks with`args` and the`PromptSubmit` and`Notification` events, and skips a hook on an event Auggie never fires, with a note.`agnostic-ai hook run` sends Auggie 1.x tool names and expands`${VAR:-default}` in a hook's command the way Auggie does. Settings`permissions.ask` ,`model` and`effort` reach Auggie's settings. See[Augment hooks](https://agnostic-ai.org/docs/targets/augment/#hooks) .
- **Cursor:**`cursor-agent` chats are listed with their folder even without an editor chat index, and a running one opens its terminal.`conversations start --tool cursor` starts a`cursor-agent` chat; Cursor editor chats are refused.

These bullets describe agnostic-ai changes checked against its own tests and reference pages on November 12, 2026; no new vendor release was audited and native behavior was not run.