Quick Start
python scripts/codex_bridge.py --cd "/path/to/project" --PROMPT "Your task"
Output: JSON with success, SESSION_ID, agent_messages, stream_file (path to the raw JSONL stream persisted line-by-line), optional stderr (codex diagnostics; lines repeated verbatim except for their timestamp are collapsed to one line with an [xN] count), and optional error.
Headless exec runs with approvals disabled by Codex itself; bridge --yolo maps to Codex's --dangerously-bypass-approvals-and-sandbox. Stdin is detached, so an approval prompt could never be answered. On Windows the PROMPT is delivered via stdin (- positional) to bypass cmd.exe quoting/length limits.
Parameters
usage: codex_bridge.py [-h] [--PROMPT PROMPT] [--cd CD]
[--sandbox {read-only,workspace-write,danger-full-access}]
[--SESSION_ID SESSION_ID] [--skip-git-repo-check]
[--return-all-messages] [--image IMAGE] [--model MODEL]
[--yolo] [--profile PROFILE] [--stream-file STREAM_FILE]
[--idle-timeout IDLE_TIMEOUT] [--ignore-user-config]
[--dangerously-bypass-hook-trust]
{mcp,plugin} ...
Codex Bridge
options:
--PROMPT PROMPT Instruction for the task to send to codex.
--cd CD Set the workspace root for codex before executing the task.
--sandbox {read-only,workspace-write,danger-full-access}
Sandbox policy for model-generated commands. Defaults to `read-only`.
--SESSION_ID SESSION_ID
Resume the specified session of the codex.
--skip-git-repo-check
Allow codex running outside a Git repository.
--return-all-messages
Return all messages (reasoning, tool calls, etc.).
--image IMAGE Attach image files to the initial prompt.
--model MODEL Model for the session (only when user explicitly requests).
--yolo Bypass approvals/sandboxing (last resort).
--profile PROFILE Load `~/.codex/<name>.config.toml` profile (only when user requests).
--stream-file STREAM_FILE
Persist raw JSONL stream path.
--idle-timeout IDLE_TIMEOUT
Kill codex if no output for N seconds (default 600; 0 disables).
--ignore-user-config Do not load `$CODEX_HOME/config.toml` (auth still uses CODEX_HOME).
--dangerously-bypass-hook-trust
Run Codex hooks without persisted hook trust. DANGEROUS.
subcommands:
mcp Thin passthrough to `codex mcp` (list/get/add/remove/login/logout).
plugin Thin passthrough to `codex plugin` (add/list/remove/marketplace).
Multi-turn Sessions
Always capture SESSION_ID from the first response for follow-up:
# Initial task
python scripts/codex_bridge.py --cd "/project" --PROMPT "Analyze auth in login.py"
# Continue with SESSION_ID
python scripts/codex_bridge.py --cd "/project" --SESSION_ID "uuid-from-response" --PROMPT "Write unit tests for that"
Config Inheritance
| Source | Inherited by default? |
|---|
~/.codex/config.toml | Yes (unless --ignore-user-config) |
Profile (--profile) | Only when explicitly passed |
Project .codex/ (trusted) | Yes, when Codex trusts the project |
Hooks / MCP / plugins / skills / AGENTS.md | Yes, via Codex runtime |
Management Passthrough
python scripts/codex_bridge.py mcp list
python scripts/codex_bridge.py plugin list
Common Patterns
Prototyping (read-only, request diffs):
python scripts/codex_bridge.py --cd "/project" --PROMPT "Generate unified diff to add logging"
Verification (codex must run tests/builds — read-only blocks them):
python scripts/codex_bridge.py --cd "/project" --sandbox workspace-write --PROMPT "Run pytest, fix the failure, output a unified diff patch"
Keep requesting patch text; audit afterwards with git status && git diff.
Debug with full trace:
python scripts/codex_bridge.py --cd "/project" --PROMPT "Debug this error" --return-all-messages
Headless hooks (vetted automation only):
python scripts/codex_bridge.py --cd "/project" --PROMPT "..." --dangerously-bypass-hook-trust