dsh-follow-up-prompts
Manifest validShows up to three follow-up instructions above the composer after each completed turn; hover shows the full text, click fills the draft, double-click or the send arrow sends it.
dsh-follow-up-prompts
English | 简体中文
A DeepSeek Harness plugin that suggests what to do next, right above the composer.
After every completed turn the plugin asks the session's own model for up to three follow-up instructions and shows them as chips above the input box:
┌──────────────────────────────────┬───┐ ┌──────────────────────────────┬───┐ ╭───╮ ╭───╮
│ Add unit tests for the parser │ ↑ │ │ Run the full test suite │ ↑ │ │ › │ │ × │
└──────────────────────────────────┴───┘ └──────────────────────────────┴───┘ ╰───╯ ╰───╯
Each suggestion is a single piece of text: what you read on the chip is exactly what gets sent. Prompts stay under 40 characters; longer ones are cut at 40 characters with an ellipsis on the chip, and the full text is shown in the hover card. Chip width follows the text, so chips are not equally wide.
- Click the text — the instruction is placed into the composer draft; edit it before sending.
- Double-click the text — sends that instruction right away, without stopping in the draft.
- Hover a chip — after a moment a card appears above it with the full instruction and the rules of use (
Click to add to the input (double-click to send) · ↑ sends now). The card is as wide as its content up to the composer width and never truncates; it ignores the pointer and disappears when you move away. - Click ↑ — the instruction is sent immediately.
- Click › / ‹ — when the chips do not fit, the row scrolls sideways. The right arrow jumps to the last suggestion, the left arrow jumps back to the first. Each arrow only appears when there really is more content in that direction; the scrollbar itself stays hidden.
- Hover, then click × — hides the row for the current turn.
Escdoes the same. While a right arrow is showing (more chips to the right), the×stays out of the way — the two share the same spot — and comes back once the row is scrolled to the end.
Requirements
- DeepSeek Harness with the
conversation.input.dockslot (any recent desktop build). - No runtime dependencies.
Install
Install the folder as a bundle (a local folder install is linked, so edits to the source take effect after a restart):
plugin_manager install_bundle target=/absolute/path/to/dsh-follow-up-prompts
Then restart DeepSeek Harness so the Host and Client halves load.
How it works
| Half | File | Responsibility |
|---|---|---|
| Host | index.js | Keeps the completed turn's user/assistant text per session, calls the model once after turn/end (reason.kind === "completed"), appends follow-up-prompts/ready to the session log. |
| Host | index.js | Folds that event into the followUpPrompts projection, which is what the client reads. |
| Client | client.js | Renders the chips in conversation.input.dock (order 30), using theme tokens only. |
Details:
- The model call reuses the session's provider/model and is capped at 600 tokens with a 20 second timeout.
- At most three suggestions are kept; the row scrolls sideways when they do not fit.
- Suggestions carry one text field (
prompt) only, so the chip text and the sent text can never diverge. - Hovering the row reveals a close button;
Escalso collapses it. The close button waits while the right arrow is showing, then returns at the end of the scroll. The row comes back on the next turn. - Subagent sessions are skipped.
- On
turn/startthe previous row is cleared and any in-flight request is aborted, so stale suggestions never flash back. - If the model returns something unparsable, the row simply does not appear.
Privacy
The turn text is sent only to the provider and route the session already uses. Nothing else leaves the machine and there is no telemetry.
Files
index.js Host half
client.js Client half
cordis.patch.yml Bundle patch
locale/en.json Display metadata (English)
locale/zh.json Display metadata (Chinese)
docs/DETAIL.md Detail page shown in the plugin list
icon.svg Plugin icon
License
MIT
Comments
Loading…
Similar plugins
by Tleon-H
After each reply, shows a gray next-prompt suggestion in the empty composer that Tab or → accepts, and ↑ in an empty composer recalls your last sent message; nothing is sent automatically.
★ 0
MIT
JavaScript
Oct 28, 2026
dsh plugin --profile web add dsh-next-promptby huaguoxingkong
One-click prompt enhancement beside the DeepSeek Harness composer: rewrites a draft in place without sending it.
★ 299
↓ 1.6k/wk
MIT
JavaScript
Nov 19, 2026
dsh plugin --profile web add dsh-prompt-rewriterAdds up to 3 clickable next-action chips under a reply; clicking one fills the input draft and never sends.
★ 0
MIT
JavaScript
dsh plugin --profile web add dsh-next-actionsby songshuhuoban
Suggests the next user reply in an empty composer after the assistant finishes, using the session model and rolling context summaries, with Tab to accept.
★ 0
MIT
TypeScript
Oct 18, 2026
dsh plugin --profile web add dsh-next-inputOne-click prompt enhancement for the DeepSeek Harness composer: a chip beside the model selector rewrites the draft into a structured prompt using the session's own model.
★ 0
MIT
JavaScript
dsh plugin --profile web add dsh-prompt-enhanceDSH composer plugin: drafts your next line after each turn; first Enter fills the input, second Enter sends.
★ 0
MIT
JavaScript
dsh plugin --profile web add dsh-saynext