@zingspark/wf-extension-search-deep
v0.1.0
Published
Deep source-diverse research bundle for wf-search: platform skills, media/video/ASR/VLM passes, and evidence scoring.
Maintainers
Readme
@zingspark/wf-extension-search-deep
Deep source-diverse research bundle for the Harness wf search runtime.
Optional package extension; never installed by default.
What it contains
This package is a pure payload bundle — no executable, no postinstall. The
wf CLI routes a research question into a plan; the agent controller loads
only the skills it needs, lazily, from this bundle.
| File | Purpose |
|---|---|
| extension.json | Extension manifest: id, capabilities, platforms |
| workflow.json | collect → understand → decide → output research stages, video policy |
| agent.json | Subagent profile and context boundaries |
| skills/manifest.json | 14 lazy-loaded skills (platform readers, media download, ASR, VLM, evidence scoring) |
| toolchain.json | Optional runtime tools (GPT Researcher, STORM, Whisper) and install roots |
| search-providers.json / source-catalog.json | Provider routing and source catalog |
| docs/ | Evidence contract, platform adapters, video research patterns |
Install
pnpm exec wf extension install wf-search-deep --project . --apply
# or directly:
pnpm add --save-dev @zingspark/[email protected]Requires the native runtime: pnpm add --save-dev @zingspark/[email protected].
Heavy optional tools (yt-dlp, gallery-dl, Whisper backends, VLM endpoints,
browser sessions) are user-provided and described in toolchain.json and the
skill manifests — nothing is downloaded at install time.
Result contract
Stable output contract: wf-research/result.v1 (see docs/evidence-contract.md).
