Expand description
Last resort: start a NEW provider process that resumes the session for one turn. Only for a session marked headless (no client holds it open). A session that has a live client is never spawned into: two writers on one transcript is how sessions get corrupted.
| Kind | Command (prompt never in argv when the CLI reads stdin) |
|---|---|
| grok/local | grok --resume <id> -p <prompt> |
| codex/local | codex exec resume <id> - (prompt on stdin) |
| kimi_code/local | kimi -S <id> -p <prompt> |
| claude_code/local | claude --resume <id> -p (prompt on stdin) |
| cursor/local | agent --resume <id> -p <prompt> |
| codex/web | codex cloud exec --env <env> <prompt> (new cloud task) |
The binary is M4A_<PROVIDER>_BIN or the vendor name on PATH.
Structs§
- Resume
Spawn Adapter - Spawns one resume turn for a headless session. Does not wait for it.
Constants§
- CODEX_
CLOUD_ ENV_ ENV - Env var naming the cloud environment for
codex cloud exec --env.
Functions§
- bin_env
- Env override for a provider binary (
M4A_CODEX_BIN, …). - default_
program - Default program name on
PATH. - resume_
command - Argv (after the program) and the stdin payload for one resume turn.