Codex CLI v0.149.0: The Multi-Session Upgrade — agents Dashboard, codex queue, and Working Directory Commands
Codex CLI v0.149.0: The Multi-Session Upgrade — agents Dashboard, codex queue, and Working Directory Commands
Released on 20 August 2026, Codex CLI v0.149.0 is the release that turns isolated sessions into a coordinated fleet.1 Three interconnected additions — the codex agents dashboard, the codex queue cross-session messaging primitive, and the /cd//pwd//cwd TUI working directory commands — address a real operational gap: running multiple concurrent agent sessions without losing the thread. This article covers all three, plus SDK reasoning effort options and codex doctor expansions introduced in the same release.
The Problem v0.149.0 Solves
Before this release, Codex CLI sessions were effectively islands. You could run multiple agents in parallel windows, but there was no first-class way to:
- See all running sessions in one place without leaving the current TUI
- Send a message from a CI job or script into an already-running session
- Redirect a session at a different directory without restarting it and losing context
The result was that multi-session patterns — coder/architect splits, parallel implementation variants, background long-running agents — required ad-hoc glue: shell scripts polling session state, copying messages by hand, or relying on subagent orchestration for tasks that didn’t justify full task-tree overhead.
v0.149.0 adds the missing glue at the CLI level.
The codex agents Dashboard
The new codex agents command launches a full-screen interactive dashboard listing all root sessions on the shared local app server.1 You can reach it in two ways:
# From a terminal
codex agents
# From inside a running Codex TUI
/agents
The default keyboard shortcut is Alt+A.2 It can be remapped via the keymap configuration, and the system disables the default when it would conflict with an existing custom binding, so you won’t silently lose a shortcut you’ve already set.
Within the dashboard, the core operations are:
| Action | Description |
|---|---|
| Search | Filter sessions by name or project — essential once you have more than a handful |
| Start | Launch a new session without leaving the dashboard |
| Open | Jump into an existing session’s TUI |
| Rename | Update a session title for clarity |
| Stop | Terminate a session directly |
Sessions are grouped by project or status, and subagent status is surfaced inline so you can see whether a session’s child agents are active without drilling in.
Configuration
Dashboard shortcut binding lives in ~/.codex/config.toml under the keymap context:
[tui.keymap.global]
open_agents = "alt-a" # default; remap as needed
[tui.keymap.agents]
# dashboard-specific bindings
search = "/"
new_task = "n"
rename = "r"
stop = "ctrl-d"
The codex queue Command
codex queue sends a message to an existing local or remote session from outside that session.1 The canonical use case is CI/CD: a pipeline job that doesn’t want to create a new agent session from scratch but needs to feed instructions into an agent that’s already holding project context.
Basic Usage
# By session name
codex queue --session my-refactor "The migration tests are green. Proceed with the schema change."
# By UUID (safer in scripts — no ambiguity)
codex queue --session "3f2b1c4a-..." "LGTM. Deploy to staging."
# Load a long prompt from a file
codex queue --session coder-backend --text-file /tmp/spec.md
# Remote session
codex queue --session my-refactor --remote wss://ci-host:7432 "Ready to merge."
Session Resolution and Validation
The resolution model is strict by design:3
- Sessions are matched by UUID or exact name. Partial or substring matches are rejected.
- If multiple sessions share the same name, the command fails rather than silently targeting one. This prevents misdirected messages in shared environments.
- Incompatible server configurations are reported explicitly — the command will not silently reroute to a different target.
- Empty messages and image attachments are rejected at validation time.
Delivery Semantics
v0.149.0 fixes three reliability issues that made the previous queuing mechanisms fragile:1
- Idle wake-up. If the target session has finished its last turn and is waiting, a queued message reliably activates it and starts a new turn.
- Mid-turn queueing. If the agent is currently processing, the message waits in queue and appears as the next user turn once the current one completes — no turn is interrupted.
- Permission restoration. Resumed or forked sessions correctly restore their permission profile before processing queued input.
Multi-Session Orchestration Patterns
sequenceDiagram
participant CI as CI Pipeline
participant A as Architect Session
participant C1 as Coder Session A
participant C2 as Coder Session B
CI->>A: codex queue --session architect "Spec complete. Delegate."
A->>C1: (subagent or codex queue) "Implement adapter pattern"
A->>C2: (subagent or codex queue) "Implement strategy pattern"
C1-->>CI: Session pushes branch
C2-->>CI: Session pushes branch
CI->>A: codex queue --session architect "Both branches ready. Pick winner."
For fork-based parallel variants (introduced in v0.148.0), codex queue pairs naturally with codex exec fork:
# Create two variants from the same base
codex exec fork --session my-task --fork-name variant-a
codex exec fork --session my-task --fork-name variant-b
# Feed different implementations to each
codex queue --session variant-a "Implement the adapter pattern"
codex queue --session variant-b "Implement the strategy pattern"
Security note: queued messages inherit the target session’s existing permission profile. No permission escalation occurs via the queue path.4
Working Directory Commands: /cd, /pwd, /cwd
Three new TUI slash commands address a mundane but frequently painful limitation: once you start a session in a directory, you’re stuck there unless you restart.1
/pwd # Print the session's current working directory
/cwd # Synonym for /pwd
/cd path # Change the session's working directory
The immediate use case is monorepos. If you start a session at the repo root while debugging a backend issue, then discover the fix requires touching the frontend package, /cd packages/frontend redirects the session without destroying the existing conversation context.
# Inside Codex TUI
/pwd
# → /home/user/my-monorepo
/cd packages/auth
# → Changed working directory to /home/user/my-monorepo/packages/auth
/cd ../api
# Relative paths work; resolves to /home/user/my-monorepo/packages/api
Note that the session’s context — the conversation history, permission profile, loaded AGENTS.md files, and active MCP servers — is preserved across a /cd. What changes is only the base directory that relative file paths and shell commands resolve against.
SDK Enhancements: max and ultra Reasoning Effort
v0.149.0 extends the SDK surface for programmatic Codex CLI usage.1 SDK callers can now pass exact CLI configuration overrides and select max or ultra reasoning effort levels:
import { CodexClient } from "@openai/codex";
const client = new CodexClient({
model: "o4-pro",
reasoningEffort: "ultra", // "low" | "medium" | "high" | "max" | "ultra"
configOverrides: {
approval_policy: "on-failure",
sandbox: { network: "off" }
}
});
ultra maps to the highest reasoning budget the selected model supports — useful for architectural tasks where token cost is secondary to correctness.5 max is equivalent to the high effort level with additional reasoning steps enabled on supported models.
Expanded codex doctor Diagnostics
codex doctor gains four new check categories in v0.149.0:1
flowchart LR
D[codex doctor] --> EP[Endpoint Protection]
D --> NP[Network / Proxy]
D --> DA[Desktop App State]
D --> UC[Update Connectivity]
D --> E[Existing checks...]
| Check | What It Validates |
|---|---|
| Endpoint protection | Confirms the selected API endpoint is reachable and authenticated |
| Network/proxy failures | Detects proxy misconfigurations that silently drop MCP or app-server connections |
| Desktop app state | Verifies the shared app server daemon is running and responsive |
| Update connectivity | Checks whether the CLI can reach the update channel (catches air-gapped environments before a session starts) |
Run codex doctor --verbose to see the raw check output. In CI environments where the desktop app is absent, codex doctor now correctly suppresses the desktop-app checks rather than reporting false failures.
What This Release Means for Workflow Design
graph TD
A[codex agents dashboard] -->|visibility| B[All sessions in one view]
C[codex queue] -->|coordination| D[CI/scripts feed running sessions]
E[/cd /pwd /cwd] -->|flexibility| F[Redirect session without restart]
B --> G[Multi-agent fleet management]
D --> G
F --> G
v0.149.0 is less about new agent capabilities and more about operational maturity. Once you routinely run three or more concurrent Codex sessions — which is normal at o4-mini throughput costs — the absence of a session overview and cross-session messaging becomes a genuine friction point. This release removes that friction without requiring subagent overhead for simple coordination tasks.
The codex queue primitive in particular opens up event-driven patterns: a CI webhook can feed test results into a running refactor session; a cron job can inject a daily summary into a long-running background agent; a separate terminal can deliver corrective input to a session that’s going off-track without interrupting its current turn.
Citations
-
OpenAI. “Release 0.149.0.” GitHub. 20 August 2026. https://github.com/openai/codex/releases/tag/rust-v0.149.0 ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7
-
openai/codex. “Add configurable shortcuts for the agents dashboard.” Pull Request #39142. GitHub. https://github.com/openai/codex/pull/39142 ↩
-
openai/codex. “Add a command to queue messages for existing sessions.” Pull Request #39092. GitHub. https://github.com/openai/codex/pull/39092 ↩
-
Vaughan, Daniel. “codex queue and Inter-Session Messaging: What v0.149.0’s New Primitive Means for Orchestration and Automation.” Codex Knowledge Base. 21 August 2026. https://codex.danielvaughan.com/2026/08/21/codex-queue-inter-session-messaging-codex-cli-v0149-orchestration-automation-agent-to-agent/ ↩
-
OpenAI. “Codex CLI 0.149.0: Task Dashboard Explained.” Get Claude Skills. 2026. https://www.getclaudeskills.com/blog/codex-cli-task-dashboard-0-149-0 ↩