What happened
Factory’s changelog identifies CLI v1.12.0 and Desktop v1.12.0 as releases dated November 14, 2026. The entry is headed “Namespaced tool names, fewer terminal redraws, and fixes across sessions, settings, and connectors.” Factory categorizes the release as applying to the CLI and app, and divides its contents between improvements and bug fixes. The entry describes changes to the Droid coding-agent product rather than a new model launch.
The most direct observability change is in the rows that record tool calls. Factory says they now show both a namespace and a capability for Droid commands, connectors, and MCP calls. In practice, a person reviewing an agent session receives a more specific label for the call than a generic tool reference. Factory describes this as a presentation change to tool rows; it does not say that v1.12.0 changes which tools a Droid can invoke or introduces a new permission model.
The release also revises Git risk treatment. A plain push or pull request now counts as medium risk, while force pushes remain high risk. Separately, revoking a service-account key now creates an audit entry for the organization. Factory says the session sidebar now hides automation sessions until the user enables the relevant filter. Together, those changes affect what reviewers see: the severity label associated with Git operations, an organization-level record of a credential event, and the default visibility of automated sessions.
The fixes concentrate on the continuity of everyday sessions and administration. After an account switch, organization settings should no longer be saved under the wrong account. A run started on a saved session ID now sends earlier conversation history with its first request. Connector settings now display a load error and Retry control rather than reporting that no connectors are available. Factory also says a terminal resize no longer blanks the transcript while it redraws, a todo list is recorded once in the session whose call wrote it, and turning telemetry off also stops onboarding events from being sent.
Several of these details matter most in long-running or multi-user setups. Session restoration determines whether a resumed agent receives the context it needs at the start of work. Connector error reporting affects whether an unavailable integration looks like an empty configuration. Account-scoped settings and telemetry controls matter to administrators who need behavior to remain consistent as people move between accounts or organizations. Factory reports these as resolved software behaviors; the changelog does not provide benchmarks or outcome measurements.
What the archive adds
New Runtime’s earlier coverage described Factory making Droid available through macOS and Windows desktop applications and a command-line interface in a November 1, 2026 note. This is a new release event in the same product line, adding interface, risk-label, audit, and reliability changes rather than restating the distribution announcement.
Trend signal
The release modestly supports agent security moving to runtime boundaries. Risk classification for Git actions and an audit entry for key revocation make operational controls more visible around an agent’s work. The clearer naming of command, connector, and MCP calls also helps identify the capability involved in an action. Factory does not establish how these controls are enforced or whether they reduce incidents, so this is a narrow operational signal rather than evidence of a broader security outcome.
Why now
The timing follows a rapid sequence of Droid releases in Factory’s November changelog. The preceding v1.11.0 entry, dated November 13, added Windows command sandboxing for supported Windows 11 hosts, while v1.12.0 arrived on November 14 with further interface, account, connector, and session refinements. That cadence suggests Factory is continuing to harden and clarify an already-shipped CLI and desktop workflow rather than reserving operational changes for infrequent major versions.
Connections to other events
An earlier New Runtime note on November 1 covered Droid’s availability through desktop applications and the CLI. This release is connected by the same distribution surfaces: Factory lists the changes for both CLI and Desktop v1.12.0. The connection is practical rather than architectural. As the product is used through more than one client, consistent session history, account handling, tool-call labels, and connector feedback become part of the operating experience across those surfaces.
Why it matters
Teams adopting coding agents need to understand not only what an agent can do, but which capability it used and how consequential a proposed action may be. Specific namespace and capability labels can make a session review less ambiguous, especially where a workflow mixes native Droid actions, connectors, and MCP tools. A clearer record is useful when an operator needs to distinguish a repository operation from an external integration call.
Risk labels and audit entries are also building blocks for operational review. Treating ordinary pushes and pull requests differently from force pushes gives the interface a more explicit distinction between common delivery work and a potentially destructive history rewrite. Recording revoked service-account keys gives administrators a trace of an access-related event. Neither feature replaces access boundaries, approval rules, branch protections, or testing. Their value is in making those surrounding controls easier to operate and investigate.
For teams already using Droid, the immediate task is modest: inspect whether the new labels map cleanly onto existing review practice. Teams should decide who reads medium-risk actions, what conditions still require human approval, and how credential-revocation records enter their incident or access-review process. The release improves signals; teams still own the policy attached to them.
What happens next
- Factory adds more explicit controls or explanations around risk classification and approvals. (3 months, 55%). Watch for: release notes describing configurable risk rules, approval behavior, or organization policy settings.
- Tool-call presentation expands further across Droid’s connector and MCP workflows. (6 months, 60%). Watch for: changelog entries that add filtering, histories, permissions, or review views for individual capabilities.
Limits and uncertainty
Factory’s changelog states the behavior changes but does not quantify adoption, time saved, error reduction, or security impact. It does not explain in the v1.12.0 entry how the medium- and high-risk labels affect approvals, whether organizations can customize the classifications, or how long key-revocation audit records are retained. The company’s release notes have not been independently tested for this Field Note. A named tool capability can improve inspection, but the entry does not show whether users can export, search, or otherwise operationalize those labels outside the session interface.
Skill to build now
Map your coding-agent actions to a short review matrix this week. Separate routine repository actions, destructive Git operations, credential changes, connector calls, and MCP calls; assign an owner and a review expectation to each. Then compare that matrix with the labels Droid presents in a real session. If you use MCP or connectors, do not rely on a generic “tool used” record: capture the namespace and capability in your own review notes or logs where possible.
Reader takeaway
Droid v1.12.0 makes selected agent actions easier to inspect: tool calls gain clearer names, ordinary pushes and pull requests are labeled medium risk, and revoked service-account keys generate audit entries. Treat these as operational signals to integrate with permissions, approvals, and testing—not as substitutes for them.
Open question
Can Factory’s risk labels be configured to match an organization’s own approval policy, and what audit history is retained for key revocations?