agent
A Claude Code settings key, read out of the shipped bundle. It was in the build from v2.1.138 and was gone by v2.1.284.
Start every session as a named subagent with its prompt, tools, and model
Anthropic's own wording. Read off All settings, captured 2026-10-22.
In the changelogs
39Releases whose published page names agent.
-
v2.1.305
Resumed workflows handle replaced and unsaved agent results more carefully
Workflow runs now track replaced journal entries, normalize saved results, and warn when an agent() result cannot be saved for resuming
found in this entry's text
-
v2.1.305
Subagent details can also be sent to a session's event stream
Saved details about a subagent can now also be sent as an event on the session's event stream, not only written to a file
found in this entry's text
-
v2.1.292
Messages sent with SendMessage now record what kind of recipient they go to
SendMessage calls shown to hooks and transcripts now include
recipient_kind, and messages to cloud and Remote Control sessions use new routingfound in this entry's text
-
v2.1.285
Workflow scripts no longer crash on an unhandled failed agent call
When a workflow script leaves a failed agent(), parallel(), pipeline() or workflow() call unhandled, Claude Code logs it and keeps running
found in this entry's text
-
v2.1.281
Agent seating can re-apply the agent system prompt
Starting a session with --agent now goes through one helper, which can also re-apply the agent's system prompt when that agent is already active
found in this entry's text
-
v2.1.281
--no-session-persistence check moved earlier and now applied immediately
--no-session-persistenceis checked earlier and takes effect at once, and--managed-settingsand setting sources are passed to child processesfound in this entry's text
-
v2.1.274
New MCP tool metadata field describing the serving MCP server
MCP tools now carry structured metadata identifying which server and source they came from, for security purposes
found in this entry's text
-
v2.1.269
Title-bar agent/tmux hint rendering split into two independent slots
The title bar's agent name and tmux hint are now computed as two separate, independent pieces
found in this entry's text
-
v2.1.269
Plugin runtime API gains a 'config' surface alongside agent/fs/store/etc.
Plugin runtime API gains a new 'config' capability alongside turn, tool, agent, fs and others
found in this entry's text
-
v2.1.267
Hooks engine gains session.compact, session.receive, ui.resolve event contracts
The plugin hooks system adds validated event types for session compaction, session receive, and UI resolution
found in this entry's text
-
v2.1.265
Hook registry gains command.describe as a core hook event
Claude Code adds a new internal hook event called command.describe
found in this entry's text
-
v2.1.260
Workflow tool: stricter schema validation for structured agent() output
agent() structured output schemas now must be valid objects, or they fail immediately.
found in this entry's text
-
v2.1.259
Per-scope row cap added to the agent-orchestration "world" engine
The workflow fact engine now caps rows per scope, emitting a runaway event and retracting the scope when it is exceeded.
found in this entry's text
-
v2.1.257
Workflow docs: model-override guidance suppressed when CLAUDE_CODE_SUBAGENT_MODEL_FORCE is set
With the forced subagent model set, the workflow authoring reference drops its model-override guidance.
found in this entry's text
-
v2.1.257
New
CLAUDE_CODE_SUBAGENT_MODEL_FORCEenv varA new env var stops per-call model overrides for subagents and workflow agents, forcing them to inherit.
found in this entry's text
-
v2.1.257
agent() workflow-script signature tightened; top-level agent option added
Workflow agent() rejects extra positional arguments and takes a top option, while the old budget options are gone.
found in this entry's text
-
v2.1.247
The
interruptible_tool_in_progressstream event is goneClients no longer get the stream event announcing that a long tool is still running and interruptible.
found in this entry's text
-
v2.1.247
Shell commands started by the main turn survive a plugin turn abort
A shell command keeps running in the background when a plugin aborts the turn that started it.
found in this entry's text
-
v2.1.246
Per-class memory caps on spawned processes, wired through but off for everything except Bash
Memory limits now reach npm, git, tmux, LSP, MCP and hooks, but everything except Bash is excluded.
found in this entry's text
-
v2.1.246
Plugins can abort the current turn with
$.turn.abortPlugins get a new turn object whose abort verb ends the running turn with its own interrupt wording.
found in this entry's text
-
v2.1.246
Cloud agent output is sanitized before use
Text coming back from a remote-isolation agent is cleaned before you see it or it's logged.
found in this entry's text
-
v2.1.242
Backgrounded Bash commands now keep a task awake
Waiting on a backgrounded shell command now keeps a task alive, controlled by a remote setting defaulting on.
found in this entry's text
-
v2.1.233
No more frontmatter warnings for reference docs alongside agents
Reference docs sitting next to your agents no longer trigger missing-frontmatter warnings.
found in this entry's text
-
v2.1.232
GitLab tokens are now stripped from logs and error output
GitLab tokens are now masked in logs, and GitHub token matching covers one more prefix.
found in this entry's text
-
v2.1.232
GitLab tokens are redacted from logs, and glab config directories are blocked in the sandbox
GitLab tokens are masked in logs and glab config directories are blocked inside the sandbox.
found in this entry's text
-
v2.1.232
Clear error when every tool an agent could use is denied
If your permissions deny every tool an agent could use, you now get a clear explanation.
found in this entry's text
-
v2.1.227
Workflow tool description does not mention the new command clamp
The Workflow tool enforces a Bash command limit but never mentions it in its description.
found in this entry's text
-
v2.1.227
Spawned agents can be clamped to a fixed set of shell commands
Spawning an agent with a bad command clamp list now fails immediately instead of running unclamped.
found in this entry's text
-
v2.1.224
Advisor entry added to Managed Agents multiagent rosters
Multiagent rosters can include one advisor entry so the main thread consults a stronger model mid-turn.
found in this entry's text
-
v2.1.224
vault_idsdocumented as create-only, and deployments can be updatedDocs say vault_ids cannot change mid-session, and deployments now have an update endpoint.
found in this entry's text
-
v2.1.223
Notice when a subagent's model is restricted
You now get a warning when a subagent's requested model gets swapped for a different one.
found in this entry's text
-
v2.1.163
Slash Command Menu Kind Lanes [In Development]
found in this entry's text
-
v2.1.163
Workflow VM Security Hardening
found in this entry's text
-
v2.1.105
Custom Subagent Status Lines
found in this entry's text
-
v2.1.69
enableWeakerNetworkIsolationSandbox Settingfound in this entry's text
-
v2.1.47
Plugin Settings from settings.json
found in this entry's text
-
v2.1.40
macOS Sandbox: TLS Certificate Verification
found in this entry's text
-
v1.0.81
Improved Telemetry
found in this entry's text
-
v1.0.62
🆕 Agent Mentions with @-syntax
found in this entry's text
First cited
0No release's published evidence quotes this name. The ledger indexes the Evidence lines of every changelog on this site, so this says nobody here has ever quoted it; it says nothing about how old the name is.
Presence across releases
2Build by build
149One row per release since the first build this name was read out of. Absent means the build was read and the name was not in it, never mined means no bundle for that release was ever archived, and a declared type or a default is only ever what that release's own bundle stated.
Read out of the published npm bundle release by release, and out of Anthropic's own documentation as this site captured it. Nothing on this page is a description anybody here wrote about what the settings key does. All settings keys.