Document long-running command auto-queue and empty-input Enter send-now#211
Conversation
- prompt-queueing.mdx: add 'Queueing during long-running commands' section (auto-queue while an agent drives a long-running command, the '(queued until the command finishes)' row suffix, at-command-end delivery, mid-command toggle override, and the new 'Default long-running command submission mode' setting with its visibility rule), plus a 'Send the next prompt with Enter' subsection covering empty-input Enter send-now and the '⏎ to send' header hint. - full-terminal-use.mdx: note that prompts queue by default while the agent drives a long-running command, linking to prompt queueing. - all-settings.mdx: add default_prompt_submission_mode and long_running_command_submission_mode under [agents.warp_agent.other]. Co-Authored-By: Oz <oz-agent@warp.dev>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
I'm starting a first review of this pull request. You can view the conversation on Warp. I completed the review and no human review was requested for this pull request. Comment Powered by Oz |
![oz-for-oss[bot]](https://avatars.githubusercontent.com/in/3450139?s=60&v=4){kind=small}
There was a problem hiding this comment.
Overview
This PR documents empty-input Enter send-now behavior and long-running command auto-queueing across the prompt queueing, Full Terminal Use, and settings reference pages. The settings reference and Enter behavior coverage look consistent, but the long-running-command section overstates two pieces of implemented behavior.
Concerns
- The LRC auto-queue section says prompts queue whenever an agent is driving an attached long-running command, but the implementation only auto-queues for commands the agent requested; user-started commands that the agent is tagged into continue to steer immediately.
- The command-finish delivery language says all suffixed prompts send immediately, but delivery preserves FIFO order and only fires LRC-queued rows at the front of the queue; a non-LRC queued row ahead of a suffixed row blocks that suffixed row until normal queue processing reaches it.
Verdict
Found: 0 critical, 2 important, 0 suggestions
Request changes
Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).
Powered by Oz
|
|
||
| ## Queueing during long-running commands | ||
|
|
||
| When an agent is driving a long-running command—a dev server, REPL, database shell, or other interactive program it has attached to via [Full Terminal Use](/agent-platform/capabilities/full-terminal-use/)—prompts you submit are queued by default instead of steering the agent mid-command. Warp sends them to the agent automatically when the command finishes. |
There was a problem hiding this comment.
| When an agent is driving a long-running command—a dev server, REPL, database shell, or other interactive program it has attached to via [Full Terminal Use](/agent-platform/capabilities/full-terminal-use/)—prompts you submit are queued by default instead of steering the agent mid-command. Warp sends them to the agent automatically when the command finishes. | |
| When an agent is controlling a long-running command it started—a dev server, REPL, database shell, or other interactive program via [Full Terminal Use](/agent-platform/capabilities/full-terminal-use/)—prompts you submit are queued by default instead of steering the agent mid-command. Warp sends them to the agent automatically when the command finishes. If you manually start a command and tag the agent in, prompts continue to steer the agent immediately. |
There was a problem hiding this comment.
Good catch — fixed in c176058. Verified against specs/QUALITY-839/TECH.md (is_agent_requested_command() gates the trigger and excludes tagged-in/user-started LRCs). The section opener now reads "a long-running command it started" and adds: "This auto-queueing applies only to commands the agent started; if you start a command yourself and tag the agent in, your prompts keep steering the agent immediately." Same clarification applied on the Full Terminal Use page. I kept the wording em-dash-free to match the docs style guide (no em dashes in instructional text) rather than using the exact suggestion.
| While the agent is in control of the command (including while it's blocked waiting for an approval): | ||
|
|
||
| * **Prompts queue instead of interrupting.** Submitting a prompt adds it to the conversation's queue and clears the input so you can keep typing. These rows show an italic *(queued until the command finishes)* suffix after the prompt text, marking that they fire when the command ends rather than when the full response finishes. | ||
| * **Queued prompts send at command end.** When the command finishes, all of the suffixed prompts are sent to the agent immediately, in queue order—even if you manually took over the command before it finished. Prompts queued other ways (`/queue`, the auto-queue toggle, or the queue default mode) are untouched and still send under the normal end-of-response rules. |
There was a problem hiding this comment.
| * **Queued prompts send at command end.** When the command finishes, all of the suffixed prompts are sent to the agent immediately, in queue order—even if you manually took over the command before it finished. Prompts queued other ways (`/queue`, the auto-queue toggle, or the queue default mode) are untouched and still send under the normal end-of-response rules. | |
| * **Queued prompts send at command end when they are next.** When the command finishes, Warp sends the long-running-command queued rows at the front of the queue in order—even if you manually took over the command before it finished. If a prompt queued another way is ahead of a suffixed row, Warp leaves that suffixed row in place so the queue order doesn't change; prompts queued other ways still send under the normal end-of-response rules. |
There was a problem hiding this comment.
Fixed in c176058. Confirmed against the spec: command-finish delivery collects only the leading LrcAutoQueue rows and rows of other origins stop the drain, so FIFO order is preserved. The bullet now reads: "Warp sends the suffixed prompts at the front of the queue, in order ... A suffixed prompt sends this way only while it stays at the front: if you reorder a prompt queued another way ahead of it, that suffixed prompt waits and sends later under the normal end-of-response rules." Kept it em-dash-free per the style guide instead of the literal suggestion.
Tighten the new long-running command and Enter send-now sections: remove em dashes from procedural/instructional text per the docs style guide, trim a few wordy phrases, replace "advertises" with neutral wording, and point the "Send now" cross-reference at the relevant subsection. Co-Authored-By: Oz <oz-agent@warp.dev>
…fy FIFO delivery Incorporates the two important review findings (verified against specs/QUALITY-839/TECH.md in warp-internal): - Auto-queue only applies to long-running commands the agent started itself, not to user-started commands the agent is tagged into. Narrowed the wording in prompt-queueing.mdx and full-terminal-use.mdx accordingly. - Command-finish delivery preserves FIFO order: only the leading suffixed rows fire when the command ends. A prompt queued another way ahead of a suffixed row blocks it until normal end-of-response processing. Rewrote the "send at command end" bullet to reflect this. Co-Authored-By: Oz <oz-agent@warp.dev>
|
/oz-review |
|
I'm re-reviewing this pull request in response to a review request. You can view the conversation on Warp. I completed the review and no human review was requested for this pull request. Comment Powered by Oz |
There was a problem hiding this comment.
Overview
This PR documents the new empty-input Enter send-now behavior for queued prompts and the default auto-queue behavior while the Warp Agent drives agent-started long-running commands. It also adds the corresponding settings reference entries for default_prompt_submission_mode and long_running_command_submission_mode.
Concerns
- No blocking concerns found in the attached diff.
- Security pass: no security-relevant findings; this is documentation-only and does not introduce code, secrets, dependencies, or configuration that changes runtime security posture.
- Spec alignment:
spec_context.mdreports that no approved or repository spec context was found for this PR.
Verdict
Found: 0 critical, 0 important, 0 suggestions
Approve
Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).
Powered by Oz
2 participants
Documents two recently shipped Warp client features.
Feature 1: Enter on an empty input sends the next queued prompt
src/content/docs/agent-platform/local-agents/interacting-with-agents/prompt-queueing.mdx:Enterwith an empty input sends the top queued row (same as its send now button), one row per press, for both prompt and shell-command rows; normalEnterbehavior when the input has text or the CLI agent Rich Input composer is open.Feature 2: Auto-queue during agent-controlled long-running commands + new setting
prompt-queueing.mdx: new "Queueing during long-running commands" section covering the default auto-queue behavior, the "(queued until the command finishes)" row suffix, at-command-end delivery in queue order, shell-command rows staying regular queued commands, the UI state (active auto-queue toggle, queue ghost text), the per-command toggle override (⌘+Shift+J/Ctrl+Shift+J), and the derived per-conversation state. Also documents the new Default long-running command submission mode setting ("Send immediately" / "Queue until command finishes", default queue), its visibility rule (only shown when Default prompt submission mode is Interrupt response), and the Command Palette entries.src/content/docs/agent-platform/capabilities/full-terminal-use.mdx: short "Prompts queue while the agent drives the command" subsection linking to the prompt-queueing section and the setting.src/content/docs/terminal/settings/all-settings.mdx: addedlong_running_command_submission_modeand the previously missingdefault_prompt_submission_modeunder[agents.warp_agent.other].Validation
npm run buildpasses (338 pages built).Conversation: https://staging.warp.dev/conversation/e0d478b8-6dce-4363-b965-404b4132f6e1
Run: https://oz.staging.warp.dev/runs/019ebca6-bdb7-781d-bc7d-f06f24140f64
This PR was generated with Oz.