dsh-landscape
在对 DeepSeek Harness 进行更改或构建之前，运行只读能力预检。
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:cyanseek/dsh-landscape#03cfcc47817e7f5eef0e3549db088713f03743e2Landscape 以 dsh-landscape 发布，当前版本为 0.3.1。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Development version
0.3.1(GitHub source; no npm release): tested locally on Windows Node.js 24.11.1 with DSH0.1.7-rc.2(2026-09-28). Agent-scoped tool discovery and fresh observation timestamps after successful unchanged scans. Model-backed end-to-end and native macOS runs have not been repeated.
简体中文 · Website · Roadmap · Contributing
Adding or building something for DeepSeek Harness? Run Landscape first.
DSH Landscape is a read-only capability preflight: describe one change in natural language and get an evidence-backed decision before you install, replace, upgrade, compose, or build.
Try needs like:
No Landscape account. No API key. No initialization. No required configuration. If runtime inspection or live discovery is unavailable, Landscape still uses its bundled snapshot and states what it could not verify.
Install the current GitHub DSH bundle into an existing profile. For reproducible deployments, replace main with a reviewed commit SHA:
dsh plugin --profile web add github:cyanseek/dsh-landscape#main
Then ask DSH normally:
Before adding browser automation, check this environment and the ecosystem. Tell me what to use, what not to build, and the safest next action.
The native tool is named dsh_landscape. Landscape itself needs no profile path or settings; --profile web belongs only to DSH's plugin installation command and should be changed to the profile you already use.
An abridged result from the bundled ecosystem snapshot:
DSH Capability Preflight
Need
Add browser automation to my DSH setup.
Current Environment
UNAVAILABLE — ecosystem analysis continued without runtime inventory
Existing Coverage
Evidence verdict: CROWDED
- titanwings/dsh-automation — installable
- titanwings/dsh-better-browser — prototype
Risks
- UNKNOWN: installed state and compatibility were not visible on this surface.
Decision
INSTALL
Do Not Build
- The mature overlapping automation surface.
Build Only
- None.
Next Action
Review the leading project's evidence and install instructions; act only if its scope fits.
The result is advice, not a mutation. Landscape does not install, enable, disable, upgrade, uninstall, edit a profile, or create a repository.
USE, INSTALL, COMPOSE, EXTEND, BUILD, WAIT, DISABLE, or INVESTIGATE.Do Not Build and Build Only boundaries for the next Agent or developer.Use once without permanent installation:
npx -y skills use cyanseek/dsh-landscape --skill dsh-landscape --agent codex
Or install the portable Skill:
npx -y skills add cyanseek/dsh-landscape --skill dsh-landscape -g -a codex -y
The Skill runs the preflight before a DSH capability mutation or implementation and reuses the host Agent for semantic review.
Pass a need directly; no mode selection is required:
npx -y github:cyanseek/dsh-landscape "Should I install browser automation for DSH?"
The compatible explicit workflows remain available:
npx -y github:cyanseek/dsh-landscape find "browser automation"
npx -y github:cyanseek/dsh-landscape analyze "Linear integration for DSH"
npx -y github:cyanseek/dsh-landscape brief "Linear integration for DSH" --format agent
npx -y github:cyanseek/dsh-landscape status --json
analyze, find, brief, their existing options, and the diagnostic status command remain supported. The package is not advertised as published to npm; the commands above run the verified GitHub source.
Normal preflight does not require this. A standalone CLI caller that specifically wants provider-backed semantic analysis can opt into any compatible endpoint:
export DSH_LANDSCAPE_API_KEY="your-key"
export DSH_LANDSCAPE_BASE_URL="https://your-provider.example/v1"
export DSH_LANDSCAPE_MODEL="your-model"
npx -y github:cyanseek/dsh-landscape analyze "Linear integration for DSH"
Credential values are never included in Landscape output.
dsh --profile web --dump-config
dsh plugin --profile web remove dsh-landscape
Preflight adds these fields without removing the v1 evidence contract:
intent, inferred from the natural-language need;environment, with detected, partial, unavailable, or not-applicable status;decision, risks, doNotBuild, buildOnly, nextAction, and limitations;verdict, recommendation, confidence, matches, covered/missing capabilities, evidence, freshness, intelligence, and provisional markers.Existing consumers can ignore the new fields. Negative findings remain conservative: incomplete or stale discovery, failed live verification, or search-only uncertainty cannot establish a genuine gap.
| Surface | Best for | Environment view |
|---|---|---|
| DSH plugin | Native preflight during a Harness session | Best-effort, read-only runtime summary |
| Agent Skill | Preflight before an Agent changes or builds capability | Host-dependent; safe fallback |
| CLI | Natural-language checks, automation, JSON, and briefs | Ecosystem-only unless supplied by a host |
| Node API | Application integration | Caller-provided or unavailable |
| Static site/API | Browser-safe snapshot exploration | Not applicable |
import { analyzeNeed, loadAliases, loadSnapshot } from 'dsh-landscape'
const [{ snapshot }, aliasData] = await Promise.all([loadSnapshot(), loadAliases()])
const result = await analyzeNeed('Should we build browser automation?', {
snapshot,
aliasData,
})
console.log(result.decision, result.nextAction)
The versioned static data API is generated under site/api/v1.
Node.js 22 or newer is required. The package has no runtime dependencies and GitHub plugin installation runs no lifecycle build scripts.
npm run release:check
npm pack --dry-run
See CONTRIBUTING.md, SECURITY.md, and the issue forms before submitting factual changes.
MIT © 2026 cyanseek