dsh-search-boost

Search boost for DeepSeek Harness (DSH): multi-engine fused search, focused page fetching, X (Twitter) search, step-mode deep research, parallel multi-agent research, and an injected proactive-search policy.

A bundle plugin for DSH that upgrades the built-in web_search / web_fetch and registers a family of search tools.

Search layers

Two layers, switched at runtime with /web_change (persisted to ~/.dsh-search-boost-layer.json):

Layer	Engines dialed	When to use
free (keyless only)	Bing + DuckDuckGo + Yahoo + Exa MCP (exa-free) — all live-probed, no API keys	Research loops, zero-cost runs, privacy-conscious use
api (default)	Same keyless legs plus Antigravity CLI (when agy is on PATH) and keyed Tavily / Brave / Exa (when keys are present)	Maximum recall; cross-engine corroboration with paid APIs

Keyless engines run in parallel so one failure never leaves you empty-handed. Fused ranking adds cross-engine co-occurrence scoring and half-life time-decay freshness.

Features

Capability	Description
Built-in web_search + web_fetch upgrade	Registers WebSearchProvider + WebFetchProvider and patches both seam configs, so built-in search/fetch keep native citation cards while the backend runs on this plugin's engine chain and Jina-first page reader
fused_search	Multi-engine fused retrieval with complexity routing, Grok-style query preprocessing (site: / OR / quotes), domain filters, cross-engine scoring, and 6h TTL cache. Active layer via /web_change; per-call layer override supported
/web_change	/web_change free → keyless pool only; /web_change api → full pool; /web_change show → current layer + engine availability
x_search	Real-time X/Twitter: posts, users, threads. With credentials: hosted xAI tool ∥ multi-engine (site:x.com), merged and deduped. Without credentials: multi-engine + oEmbed full text (~2s), guest GraphQL user profiles, oEmbed threads. /x-login / /x-logout toggle the official path
/x-login	Import xAI credentials into ~/.dsh-search-boost-xauth.json (from ~/.grok/auth.json or -k <XAI_API_KEY>). OIDC auto-refresh; grok CLI login untouched
/x-logout	Remove /x-login credentials; x_search falls back to the credential-free chain
fetch_page	Jina Reader + local HTML fallback + focus topic extraction + 24h cache
deep_research	Step-mode deep research: complex fused search + coverage analysis + gaps + suggested queries, driven by the main agent in rounds
research_parallel	Sub-query decomposition → native DSH subagents in parallel → merged sources
search_stats	Cache / tier / engine availability / x_search credential audit
Search policy	Injected via systemPrompt.section: verify time-sensitive facts, route X content to x_search, prefer free engines

Installation (bundle — recommended)

From npm (recommended):

dsh plugin --profile web add dsh-search-boost          # latest
dsh plugin --profile web add [email protected]    # pin a version
dsh plugin --profile web update dsh-search-boost       # update to a newer release

--profile <name> is required (web is the standard web-UI profile). DSH uses pnpm to resolve the package, auto-wires the dsh.bundle.patch layer, and adds it to the profile's bundle list — no manual config editing. Restart dsh --profile web.

From source (development):

dsh plugin --profile web add github:Mr-remon219/dsh-search-boost
dsh plugin --profile web add git+file:///path/to/repo

Or use the install script (syntax check → key setup → install → verification):

.\install.ps1          # Windows
./install.sh           # Linux / macOS

Verify:

dsh --profile web --dump-config   # web.searchProvider → dsh-search-boost
dsh --profile web

Troubleshooting: missing dsh or pnpm

If you run DSH via npx @deepseek-ai/dsh web, there is no global dsh binary — install globally or use npx directly:

npm install -g @deepseek-ai/dsh
# or:
npx --yes @deepseek-ai/dsh plugin --profile web add dsh-search-boost

dsh plugin needs pnpm (npm install -g pnpm or corepack).

Alternative: session-level dynamic plugin (plugin-host.js)

plugin-host.js is a single-file dynamic plugin for per-session boosts. It does not replace built-in web_search. The bundle form (above) is recommended for deployment-level integration.

Configuration (API keys)

The published bundle contains no secrets. Keys load from:

1. ~/.dsh-search-boost-keys.json or ./.search-boost-keys.json:

{ "tavily": "tvly-...", "exa": "...", "brave": "..." }

1. Environment: TAVILY_API_KEY / EXA_API_KEY / BRAVE_API_KEY

Engines without a key are dropped from the fan-out automatically.

Free layer needs zero configuration. Bing, DuckDuckGo, Yahoo, and Exa MCP (exa-free) are keyless and run in parallel. Antigravity CLI (agy) is optional and only joins in the api layer (medium/complex tiers) when installed and signed in.

x_search credentials (optional)

Command	Effect
/x-login	Import grok login → official hosted x_search path (SuperGrok / X Premium+ tier)
/x-login -k <XAI_API_KEY>	Same via console.x.ai API key
/x-login status	Show credential chain
/x-logout	Disable official path; fallback chain still works

~/.grok/auth.json is never auto-consumed. Without /x-login or XAI_API_KEY, x_search uses only the credential-free chain.

Verified benchmarks (2026-08, Windows)

Free-layer engine probe (12 queries × 9 candidates)

Live benchmark (node scripts/engine-benchmark.mjs); full report in scripts/engine-benchmark-report.json.

Engine	Success	Avg latency	Verdict
bing	100%	~2.0s	✅ free layer
ddg	100%	~2.2s	✅ free layer
yahoo	100%	~2.3s	✅ free layer (new in v0.1.2)
exa-free	100%	~4.2s	✅ free layer
antigravity	92%	~27s	api layer only (slow, needs agy CLI)
brave-html / mojeek / searx	0%	—	rejected (429, blocked, or unreachable)

Integration scenarios

Scenario	Result
free layer fused_search	5 hits in ~1.3–3.0s; cross-engine corroboration (e.g. yahoo+exa-free → score 3.29 on rust release notes)
x_search (no credentials)	keyword via multi-engine + oEmbed; user via guest GraphQL (@NASA ~2s); thread via oEmbed
SSRF vs Clash TUN fake-ip	Literal 198.18/15 blocked; TUN-routed hostnames allowed (DSH_SEARCH_ALLOW_TUN_FAKEIP=0 to disable)
Headless web_search	✓ free Bing/DDG/Yahoo/Exa chain
Unit + E2E tests	57/57 unit tests; 14/14 black-box E2E

Architecture notes

• Runs in the host process (Node fetch / child_process directly).
• HTML scrapers (Bing / DDG / Yahoo) use IPv4-forced fetch — undici on Windows defaults to IPv6-first DNS, which intermittently times out against search hosts.
• X search: official path (lib/xsearch.js) → credential-free chain (lib/xfallback.js: multi-engine + oEmbed + guest GraphQL) with sync preflight when no credentials.

Files

index.js                    — bundle entry (providers + tools + commands + policy)
lib/engines.js              — engine registry (bing / ddg / yahoo / exa-free / …)
lib/exa-free.js             — Exa MCP keyless engine
lib/layer.js                — free/api layer state (/web_change)
lib/fusion.js               — fusion scoring, tier tables, cache
lib/fetch.js                — Jina Reader + focus extraction
lib/xauth.js / xsearch.js / xfallback.js — x_search credential + official + fallback paths
lib/research.js             — deep_research + research_parallel
lib/policy.js               — proactive-search policy
cordis.patch.yml            — DSH patch layer manifest
scripts/engine-benchmark.mjs — maintainer: live free-engine probe

Publishing (maintainers)

npm test
npm publish   # prepublishOnly: syntax + tests + clean git tree

License

MIT

Friends

• Linux.do — open-source developer community