lin-answer-now

answer-now

English | 中文

A DeepSeek Harness plugin that pushes the agent to deliver its answer as soon as it has one, instead of burning your time on extra queries.

What it does

1. Injects instructions into the system prompt — stop querying the moment you can answer, never repeat a query, prefer one decisive query over several narrow ones, no filler tool calls.
2. Counts tool calls per agent per turn — and when the count crosses a threshold, injects an escalating reminder into the agent's inbox ("answer now instead of querying further" → "stop calling tools and produce your final answer").
3. Resets on new user messages, so each of your turns starts fresh.

It is advisory-only: it never blocks or rejects a tool call, so it can't break a legitimate workflow.

Install

From npm (prebuilt, no build step):

dsh plugin --profile desktop add lin-answer-now

Or from GitHub source:

dsh plugin --profile desktop add https://github.com/lin1213812138/lin-answer-now

Then restart dsh web (or reload the page) and you're done.

Configuration

All values are optional; the defaults are sensible.

- id: answer-now
  name: lin-answer-now
  config:
    thresholds: [4, 7, 10]        # remind at these tool-call counts, escalating each time
    include: []                   # only watch these tools (empty = all)
    exclude: ["answer_now"]       # never count these tools
    quietTools: ["todo_write", "todo_read", "todo_write_many", "schedule_list"]
    cooldownSteps: 1              # tool calls to skip between reminders
    reminderSummary: "answer now" # summary text on the reminder message
    resetOnUserMessage: true      # reset the counter when you send a new message

Requirements

DSH ≥ 0.2.0.

License

MIT