From knowledge-patch
Provides reference information for OpenAI Codex 0.144.0, covering CLI, app, security, MCP, and session controls. Useful for adapting to Codex compatibility changes.
How this skill is triggered — by the user, by Claude, or both
Slash command
/knowledge-patch:codex-knowledge-patchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Baseline: before the public Codex CLI launch, with no stable CLI, web, IDE extension, desktop app, or unified product changelog; coverage extends through `codex-app` (2026-02-02), `0.144.0` (2026-07-09), `cli-overview-and-features` (2026-07-11), and `cli-reference-and-commands` (2026-07-11).
Baseline: before the public Codex CLI launch, with no stable CLI, web, IDE extension, desktop app, or unified product changelog; coverage extends through codex-app (2026-02-02), 0.144.0 (2026-07-09), cli-overview-and-features (2026-07-11), and cli-reference-and-commands (2026-07-11).
| Reference | Topics |
|---|---|
| app-and-collaboration.md | Desktop app, worktrees, shared skills, scheduled automations, personality, credits, Code Mode, image generation |
| security-and-configuration.md | Invocation overrides, approval and sandbox presets, feature flags, search, login, policies, Windows behavior, diagnostics |
| remote-mcp-and-connectors.md | Remote TUI and WebSocket authentication, MCP administration and authentication, hosted connectors, proxied Responses WebSockets |
| cli-execution-and-cloud.md | Scripted execution, structured output, resume and fork, cloud tasks, plugin marketplaces, repository workflows |
| tui-and-session-controls.md | Steering and editing keys, themes, planning, Fast mode, status UI, background terminals, conversation and agent controls |
on-failure approval policyUse one of the supported approval policies:
codex --ask-for-approval untrusted
codex --ask-for-approval on-request
codex --ask-for-approval never
on-failure is deprecated. --full-auto combines workspace-write with
on-request. --yolo bypasses both approval prompts and sandboxing; use it
only inside an externally isolated environment.
Since 0.144.0, Code Mode defaults to hosted mode. Every approval request
causes an elicitation pause. The writes app-approval mode permits declared
read-only actions without prompting and requires approval for writes.
Since 0.144.0, MCP tools may request interactive authentication by default;
an experimental opt-in is no longer required. App-server hosts may supply
Codex authentication at runtime, hosted login may redirect success to a hosted
page, and long-running sessions refresh expired authentication for the hosted
codex_apps connector.
Since 0.144.0, image-generation requests use the image-generation extension
by default. From the CLI, attach reference images with --image and quote the
literal $imagegen name in shell prompts.
codex '$imagegen Create a transparent 64px application icon'
codex --image existing.png '$imagegen Extend this banner to a 3:1 aspect ratio'
Select a sandbox and approval policy independently:
codex --sandbox workspace-write --ask-for-approval on-request
Sandbox values are read-only, workspace-write, and danger-full-access.
The desktop app uses the same native configurable sandbox as the CLI. By
default, app work is limited to the active folder or branch and cached web
search; network access and other elevated commands need permission. Project or
team rules may automatically allow named commands to run elevated.
On Windows since 0.144.0, sandboxed sessions may delete files inside writable
roots and access the managed primary runtime. In a native Windows TUI,
/sandbox-add-read-dir C:\absolute\path adds an existing absolute read path for
the rest of the session.
Read security-and-configuration.md before changing approval, sandbox, authentication, policy, or search settings.
Use codex exec or its codex e alias for non-interactive work. Supply the
prompt as an argument or use - to read it from stdin.
printf '%s\n' 'Inventory the package licenses' | \
codex exec - --json --output-schema result.schema.json \
--output-last-message final.txt
--json emits JSONL state-change events.--output-schema constrains the final response with JSON Schema.--output-last-message writes the final natural-language response separately.--ephemeral avoids saving rollout files.--skip-git-repo-check permits a workspace outside a Git repository.Resume the latest scripted session scoped to the current directory with:
codex exec resume --last "Continue with the next package"
Use --all to include sessions from other directories. resume continues the
same session; fork creates a new thread while preserving the source transcript.
Let an app server own the workspace and execute commands while another machine runs the TUI:
# Workspace host
codex app-server --listen ws://127.0.0.1:4500
# TUI through a forwarded loopback port
codex --remote ws://127.0.0.1:4500
Pass an explicit ws:// or wss:// URL. Use unauthenticated plain WebSockets
only on loopback or through an SSH-forwarded loopback connection. For a network
listener, use capability-token authentication over wss://; the client refuses
to send a token over non-loopback ws://.
codex app-server --listen ws://0.0.0.0:4500 \
--ws-auth capability-token \
--ws-token-file "$HOME/.codex/codex-app-server-token"
export CODEX_REMOTE_AUTH_TOKEN="$(ssh devbox 'cat ~/.codex/codex-app-server-token')"
codex --remote wss://devbox.example.com:4500 \
--remote-auth-token-env CODEX_REMOTE_AUTH_TOKEN
For signed bearer authentication, use --ws-auth signed-bearer-token with
--ws-shared-secret-file. Tokens are HS256 JWTs with a required exp claim;
the shared secret must contain at least 32 bytes. Read
remote-mcp-and-connectors.md for the
remaining claim-validation rules and MCP setup.
Global options include --profile, --oss, -C, repeatable --add-dir,
repeatable --enable and --disable, and -c key=value. When invoking a
subcommand, place its global options after the subcommand name. -c parses a
JSON value when possible and otherwise preserves a string.
codex --profile work -C ./repo --add-dir ../shared \
--sandbox workspace-write --ask-for-approval on-request
codex exec --profile work --enable unified_exec "Run the checks"
--oss selects a local open-source provider backed by a running Ollama
instance. One-run feature overrides do not replace persisted feature settings.
codex features list
codex features enable unified_exec
codex features disable shell_snapshot
Feature changes persist in ~/.codex/config.toml; with --profile, they are
stored in that profile. --search selects live search for one run, while
web_search = "live" and web_search = "disabled" persist the choice. A
full-access sandbox defaults to live results, and JSON execution reports
searches as web_search events.
codex cloud opens a task picker. Submit directly with an environment ID and
request one to four independent attempts:
codex cloud exec --env ENV_ID --attempts 3 "Summarize open bugs"
codex cloud list --env ENV_ID --limit 20 --json
codex apply TASK_ID
Submission failures return nonzero. Cloud listing supports environment filters,
cursor pagination, a limit from 1 through 20, and JSON containing tasks plus
an optional next cursor. codex apply (alias codex a) applies the latest task
diff, reports patched files, and exits nonzero if git apply conflicts.
Enter during a turn to inject text into it; press Tab to queue input,
slash commands, or ! shell commands for the next turn.Ctrl+G with VISUAL or EDITOR for long prompts and Ctrl+R for
history search. Up and Down restore drafts, including image placeholders./copy or Ctrl+O to copy only the latest completed output.Esc twice to edit the prior user message;
continue pressing it to walk backward, then submit to fork from that point./clear for a fresh conversation and cleared view, /new for a fresh
conversation without clearing the terminal, and Ctrl+L to clear only the
display./compact to replace older turns with a concise context-preserving
summary, /resume to select a saved transcript, and /fork to clone the
current conversation under a new thread ID.Read tui-and-session-controls.md for planning, Fast mode, themes, status layout, background terminals, apps, plugins, and spawned-agent navigation.
npx claudepluginhub nevaberry/nevaberry-plugins --plugin knowledge-patchRuns OpenAI Codex CLI for code analysis, refactoring, and automated editing. Handles model selection, sandbox mode, session resumption, and error handling.
Runs Codex CLI for code analysis, refactoring, and automated editing. Supports model selection, sandbox modes, and session resumption.
Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, refactoring, or automated editing