shujiTech/dsh-plugin-wepre
DeepSeek Harness plugin: publish single-screen content cards to WePre Next from a dsh agent session
dsh-plugin-wepre lets a dsh agent publish single-screen content cards to WePre Next — a platform that pushes offline, single-viewport cards to users while they wait on AI generations — so a session goes from 'write me a card' to 'it's live' in one conversation. Five model-facing tools run the loop: wepre_request_code (send a one-time SMS/email login code), wepre_login (redeem the code; the session cookie persists locally), wepre_whoami (login state), wepre_publish (upload an index.html/.zip card through the server-side QA gate) and wepre_qa_report (fetch the full per-viewport QA report). The tools' descriptions teach WePre's content rules (single-screen layout, no network calls/external resources, the wepre-next:execute postMessage contract) and the QA fix loop: on QA_GATE_FAILED the agent reads the structured per-viewport issues, fixes the card and republishes with the same contentId — no human in the loop. A local pre-flight scan flags banned tokens (fetch, localStorage, Worker, overflow:auto) and external links as non-blocking warnings.
Install
dsh plugin --profile web add github:shujiTech/dsh-plugin-wepreGitHub install per README (EN primary with zh edition); npm dsh-plugin-wepre 404 verified 2026-09-02. The README's first install block (dsh plugin --profile web add dsh-plugin-wepre) assumes the registry form, and its git-host form is explicitly documented as working with no build step and no prepare script: dsh plugin --profile web add github:shujiTech/dsh-plugin-wepre. Verify the layer composed (dsh --profile web --dump-config shows a '# == dsh-plugin-wepre' layer), then start dsh --profile web. Plain ESM JavaScript, no build step; MIT.
Compatibility
DSH web profile; publishes to WePre Next (endpoint default https://wepre.cn/next-test); auth via token config field, WEPRE_PUBLISH_TOKEN env, or a wepre_login session persisted to ~/.dsh/wepre-session.json (0600).
Details
- Repo: shujiTech/dsh-plugin-wepre
- Category: Agent Capabilities
- Stars: 1
- Version: git github:shujiTech/dsh-plugin-wepre
- Last push: 2026-08-13
- First seen: 2026-08-13
Recent updates
5 wepre_* tools (code/login/whoami/publish/qa_report); server-side QA gate with structured per-viewport issues; autonomous fix-and-republish loop; local pre-flight warnings; session-file auth.
FAQ
- What can the agent do end to end?
- Log in with a one-time code, publish an index.html or .zip card through the QA gate, read the structured per-viewport QA report, fix issues and republish with the same contentId — no human in the loop.
- How is authentication handled?
- Credentials resolve in order: the token config field, the WEPRE_PUBLISH_TOKEN environment variable, or the stored login session created by wepre_login (persisted to ~/.dsh/wepre-session.json, mode 0600).
- What content rules does the agent learn?
- Single-screen layout, no network calls, no external resources, and the wepre-next:execute postMessage contract; a local pre-flight scan warns on banned tokens and external links before publish.