Live Settings, Durable Work, and Broader Pull Request Watches
0.7.0Kiro Crew 0.7.0 makes day-to-day work more durable and easier to control. Most settings apply as soon as you save them, accepted background work survives Gateway restarts, and updates wait for active work to finish. Long chats gain a turn minimap, clearer tool-call titles, and direct model and reasoning controls, while pull request watches expand to GitLab, Bitbucket Cloud, and Azure DevOps Services. Windows gets isolated pods and a faster Gateway start, and Preview users can choose more agent backends, let Jev route individual turns, or run remote crews on AWS Fargate.
Most settings now reach the running Gateway as soon as you save config.json or change a value in Settings. Fields that still need a restart identify themselves in the UI; the Slack slash-command field is the current boot-only setting and carries a Changes need a restart badge.
A turn minimap in the left gutter of a desktop chat shows one marker per loaded turn, with hover previews, click-to-jump navigation, and an end cap that loads older history. The model chip now puts model and reasoning-effort choices in one place, and Selectable Models controls which options appear. When the agent is working, choose Steer to redirect the current turn or Queue to wait for the next one; Cmd or Ctrl plus Enter performs the other action for one send.
Side Chat opens from /side or /btw for an ephemeral question beside the main conversation. On sessions using the Kiro CLI backend, it can run read-only lookups without approval; changes and MCP tools are refused. Tables can be copied as Markdown or CSV, and long code blocks repeat their actions at the bottom.
Accepted subagent work, Task Runner steps, and workflow calls now enter a durable queue before they run. A Gateway restart restores queued work instead of dropping it, and the System page shows the active queue and the concurrency capacity currently in force.
For long-horizon work, the optional Crew log records an append-only session trail and powers a durable session ledger across context compaction. Set KIROCREW_CREW_LOG=1 in ~/.kiro/crew/.env and restart the Gateway to enable its status, usage, timeline, tools, and approvals views. The kirocrew-crew-log MCP server provides read-only access for agents.
The structured pull request watch now accepts GitHub, GitLab, Bitbucket Cloud, and Azure DevOps Services URLs. GitLab uses glab; Azure DevOps uses az with the azure-devops extension and either an az login session or AZURE_DEVOPS_EXT_PAT; private Bitbucket pull requests can use BITBUCKET_EMAIL and BITBUCKET_API_TOKEN.
Azure DevOps Server and Bitbucket Data Center are not supported. Azure DevOps and private Bitbucket watches must start from a dashboard conversation. Stop and inspect actions now work across plain timer loops, pull request watches, and workflow watches, including from Webex.
OpenCode and goose join Kiro CLI, Claude Code, Codex, and KAS in the Agent Backend selector. Crew verifies each backend's permission path before the first prompt and shows missing installation components in the selector. Codex and OpenCode sessions can compact, and Crew's own tools reach OpenCode and goose sessions.
Pi is also available when both pi and pi-acp are installed. Pi sessions do not carry Kiro Crew's MCP tools, so choose another backend when you need Crew subagents, scheduling, memory, or other MCP capabilities. kirocrew doctor reports this limitation.
Enable Decisions (Jev) under Settings > Developer > Feature Previews, then choose Auto (Jev) in a chat's model picker to classify each message as simple, medium, or complex and route it through decisions.model_route. Choose Auto (Jev) on the busy send button to let Jev decide whether a mid-turn message should steer or queue. Each decision appears on the message it shaped, with feedback controls.
The Preview is opt-in. Model routing and busy-send decisions start only when you select Auto (Jev) in the corresponding control. The settings card explains what content can be sent to the configured provider; separate consent switches govern tool arguments, compaction inputs, and recalled-memory snippets.
AWS Fargate joins SSH and AWS Systems Manager Session Manager (SSM) as a remote-crew connection method. Crew reaches the task through SSM with no inbound network rule, using the AWS CLI and Session Manager plugin on your laptop. Tasks live for six hours by default and the runtime admits at most ten at once. Change the lifetime with fargate.task_ttl_seconds in cloud.json; the ten-task ceiling is fixed.
The Gateway now checks for updates at startup and every twelve hours. An automatic update pauses new work and waits for active turns, scheduled runs, subagents, Task Runner steps, and workflows to become idle before applying, then restarts the Gateway. If a source update needs a newer Python than the current virtual environment provides, Crew refuses before moving the checkout and names the interpreter that needs attention.
Managed installs can also arm an update from the update and What's New dialogs, using the same time-limited host approval shown under Settings > About.
The isolated-pod workflow now runs on Windows through Task Scheduler. kirocrew pod up, down, ls, status, token, url, logs, prune, and provision cover the pod lifecycle for a Kiro Crew source worktree. pod api remains unavailable on Windows because the private request channel requires a Unix-domain socket.
With Dev Fleet enabled under Apps > Library, an agent can start or stop a worktree pod and receive its address, port, and short-lived token without leaving the chat.
skills.max_triggered now defaults to 0. Sessions receive a short skill index and agents use skill_search to load relevant instructions, instead of injecting every trigger match automatically. Set a positive value to allow up to that many matched skills to inject per message.