studyzy/dsh-suggest-prompt
dsh-plugin suggest next prompt
dsh-suggest-prompt is a single-bundle 'next-prompt suggester': after each agent turn completes (turn/end, reason=completed), it makes one bounded auxiliary LLM call and writes a suggested next prompt into the session log; the browser half renders it as a light ghost placeholder inside the input box, and pressing Tab (default) adopts it into the draft — Claude Code style. It inherits the main request's last routed provider/model when none is configured (no extra setup), runs suggestions with reasoningEffort off so they cost almost nothing, and degrades gracefully when the model doesn't support the parameter.
Install
dsh plugin --profile web add git@github.com:studyzy/dsh-suggest-prompt.gitnpm @studyzy/dsh-suggest-prompt 1.0.1 verified 2026-09-01 (maintainer studyzy; bilingual README, repo studyzy/dsh-suggest-prompt). README recommends the GitHub install dsh plugin --profile web add git@github.com:studyzy/dsh-suggest-prompt.git (or the https form); the npm form dsh plugin --profile web add @studyzy/dsh-suggest-prompt works once published. Git-hosted installs run a prepare build — pnpm >= 10 may block it; if add errors, allow the package key pnpm prints, or use the local path dsh plugin --profile web add /path/to/dsh-suggest-prompt.
Compatibility
DSH web profile; optional provider/model override for suggestion generation; reasoningEffort: off default.
Details
- Repo: studyzy/dsh-suggest-prompt
- Category: Agent Capabilities
- Stars: 2
- Version: git install (npm @studyzy/dsh-suggest-prompt 1.0.1)
- Last push: 2026-08-21
- First seen: 2026-08-13
Recent updates
Ghost next-prompt suggestion in composer; Tab-to-adopt; inherits last route; reasoning off; bounded single call per turn.
FAQ
- Do I need to configure a model for suggestions?
- No — unconfigured, it inherits the main request's most recently recorded routing; you can also pin any route explicitly.
- How do I accept a suggestion?
- Press Tab (default shortcut) to fill the suggested prompt into the draft.
- Does it cost reasoning tokens?
- No — suggestions run with reasoningEffort off by default; unsupported models auto-drop the parameter and retry once.
Alternatives
lql341/dsh-scnet · dylan121322/llm-adaptive · superboy911/dsh-model-router