-
Notifications
You must be signed in to change notification settings - Fork 0
Resuming a Build
ctrl+r on an executed version in the picker starts the agent session that
built it back up. You are back in the conversation that wrote the code, instead
of explaining the code to a stranger.
planx # → for the plan's versions, ctrl+r on the green oneplanx execute writes four things onto the version at the moment the agent
picks it up:
| Recorded | Where it comes from |
|---|---|
| The session id |
--session-id, passed by the executing agent |
| The agent |
--agent, or PlanX walking its own process tree |
| Its launch line | The same process walk |
| The plan's directory | The directory the plan was captured in |
The process walk is why the agent does not have to be told: how the tab was
started is a fact about PlanX's own parents, not something the agent knows about
itself. PlanX reads ps up to five hops, past shells and runtimes like node
or npx, and stops at claude or codex.
The recorded launch line, verbatim, minus the flags that name a session — those would collide with the selector PlanX is adding.
| Agent | The command |
|---|---|
| Claude Code | claude <recorded flags> --resume <session-id> |
| Codex | codex <recorded flags> resume <session-id> |
Codex takes resume as a subcommand, so its flags go in front of it.
It runs in the plan's own directory. If that directory is gone, PlanX says so and runs where you are instead.
The whole command is printed before it runs, flags included, so what was started and what it was granted is on the scrollback above the agent's first frame.
Resuming sends the agent nothing. The first thing it hears is whatever you type next.
That is deliberate: re-sending /planx execute would have the agent re-read the
plan and redo work it has already done. You are resuming the conversation that
implemented the plan, and what happens next is yours.
ctrl+r appears only on a row that can produce a launch.
- No session id recorded. The agent that built it never passed one — a version built before this feature existed, or by an agent that names no session.
-
An agent PlanX cannot start. Only
claudeandcodexare launched directly; anything else is a shell in between, and reviving a session with a guess is not something PlanX attempts. -
A plan row with two builds behind it. The plan row offers
ctrl+ronly where exactly one of its versions qualifies. Press→and pick the version.
Without the key, resume the session with your agent's own command — claude --resume <id> or codex resume <id> — or simply start a new session and hand
it the plan with /planx execute <id> v<n>.
If the agent is not on your PATH, PlanX says so and prints the line it was
going to run, under Resume it yourself. Nothing is lost — the command is
there to copy once the binary is back.
A version that was executed is tagged executed and painted green in the
picker. Its plan row is green too, until the plan gains a version newer than the
one that was built: what was executed is no longer the plan, so the plan row
goes back to normal and the green stays on the child row, which is where the
history is.
Start here
- Installation
- The skill
- Why review an agent's plan
- Plan mode vs. a reviewable plan
- Planning with Claude Code
- Planning with Codex
Using PlanX
Reference
Ask