modsearch

Original：🇺🇸 English
Translated

Bridge web search capability for LLM workflows. Use when user asks for latest info, external facts, or source links and the active model/toolchain lacks direct search ability.

4installs
Sourceliustack/modsearch
Added on

NPX Install

npx skill4agent add liustack/modsearch modsearch

Tags

Translated version includes tags in frontmatter
llm-workflowweb-searchgemini-clitoolchain-extensionstructured-search

SKILL.md Content

View Translation Comparison →

ModSearch - Search Bridge Skill

Use this skill when:
  • User asks for web search, latest updates, or source-backed answers
  • Current model/toolchain cannot perform direct search reliably
  • You need structured search results before downstream summarization/fetch

Prerequisites

bash
modsearch --version
If using the default provider (
gemini-cli
), ensure Gemini CLI is installed and authenticated:
bash
gemini --version
If missing:
bash
npm install -g @google/gemini-cli
gemini
Or run directly with 
npx
(no global install required):
bash
npx @google/gemini-cli          # launch interactive auth

Command

bash
modsearch -q "<query>"
# or via npx
npx @liustack/modsearch -q "<query>"
Optional:
bash
modsearch -q "<query>" -o <output-json-path> -p <provider> -m <model> --max-results <n> --prompt "<extra constraints>"

Workflow

  1. Run 
    modsearch
    with the user query.
  2. Parse JSON output.
  3. Use 
    summary
    and 
    items
    as evidence for reasoning or for selecting URLs to fetch.
  4. If 
    uncertainty
    is non-empty, explicitly communicate ambiguity to the user.

Output Contract

  • summary
    : high-level synthesis of search findings
  • items
    : normalized list (
    title
    , 
    url
    , 
    snippet
    , 
    source
    , 
    published_at
    , 
    relevance
    )
  • uncertainty
    : caveats, missing data, or confidence risks
Detailed schema: 
references/output-schema.md

Failure Handling

  • If provider command fails (missing auth, quota, network, binary not found), report exact error and suggest provider setup checks.
  • If JSON is partially malformed, keep 
    rawText
    and continue with best-effort extraction.

Implementation Note

v1 uses Gemini CLI as the default provider (
gemini -p
with JSON output mode).
The architecture is provider-extensible and can support any search-capable model or service provider in future versions.