batch · open-rua — Claude Code prompt: what it does, how to install
The batch prompt for Claude Code groups independent reads and perception queries in one model response. Physical actions stay ordered, with a pause where their result needs checking.
MIT ★ 488 updated 2026-11-15
.claude/commands/batch.mdWhat batch does
The prompt batches tool calls whose inputs are already known. Instead of spending a separate model turn on each query, the agent submits several calls in one response. A request that depends on an image, a snapshot_id, or a skill result waits until that result arrives.
Physical calls use one shared queue, so their planned order matters. The batch ends where arrival, contact, or release needs verification; speculative corrections and duplicate moves are left out. Multiple targets in one call request coordinated arms, while separate calls remain distinct actions.
To install it in Claude Code, save the prompt as .claude/commands/batch.md.
What it helps with
- group independent tool queries in one model response
- wait for an image or skill result before dependent work
- keep physical actions in their intended order
- pause a route where arrival or contact needs checking
Example prompts for the agent
- Batch the independent reads; all their inputs are ready.
- Get the image first, then plan the actions that depend on it.
- Follow the route in order and stop before checking arrival.
How to install batch
Manually: put the files in place
.claude/commands/batch.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about batch
How do I install batch in Claude Code?
Save the prompt to .claude/commands/batch.md.
What does batch do?
It groups independent tool calls in one model response and sets an order for physical actions. Dependent queries wait for their required results.
Which queries can batch combine?
Only independent queries with known inputs. A list of calls cannot supply data that has not arrived.
Source text
batch.md · MIT
Batch independent reads and perception queries when their inputs are already known. Resource scheduling determines whether calls can overlap. Receive a required image, snapshot_id or skill result before planning work that depends on it. Batch independent queries rather than spending a separate decision round on each one. Here, a batch means multiple tool calls in one model response, not one call per response followed by a final wait. Choose all already-determined calls before submitting the batch. Follow the runtime dispatch and resource ordering guarantees; a list of calls does not create missing data dependencies. Physical calls share one exclusive episode queue. Submit a preplanned route in its intended order only when the next segment remains justified without an intermediate result. End the batch where arrival, attachment, contact effect or release support must be checked. Do not queue speculative corrections or duplicate moves. Multiple targets within one call request coordinated arms; separate calls remain separate physical actions.
© Open-RUA · MIT · Open-RUA/uni-robot-agent
Similar prompts
All prompts for Agent setup: memory, context, orchestration →batch by other authors comes first.