返回插件市场

dsh-landscape

Agent 与工作流

cyanseek/dsh-landscape

面向 Agent 的 DeepSeek Harness 插件智能:验证现有插件、识别缺失能力并生成可构建的简报。用于安装、替换、升级、组合或构建决策的只读能力预检。

  • agent-skills
  • codex
  • deepseek-harness
  • dsh-plugin
  • ecosystem
  • gap-analysis
  • plugin-discovery
GitHub Stars
7GitHub
浏览量
0DSH Plugin Hub
Forks
0GitHub
开放问题
0GitHub Issues
Manifest 版本
0.3.0dsh-landscape
最近推送
2026年8月23日GitHub
许可证
MITJavaScript
插件类型
Host运行于 DSH Host

README

查看源文件

DSH Landscape

简体中文 · Website · Roadmap · Contributing

CI Node.js >= 22 License: MIT

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:

  • “Should I install browser automation, or is it already covered?”
  • “Compare the GitHub integrations that could fit this DSH setup.”
  • “Can I replace my current search plugin without losing capabilities?”
  • “Before we build a Linear integration, show what not to duplicate.”

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.

Quick start

Install the pinned, reproducible DSH bundle into an existing profile:

dsh plugin --profile web add github:cyanseek/dsh-landscape#2d3570aadbbd291dbfc58e2484e287bd14fa92e0

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.

What a preflight looks like

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.

Why run it first?

  • See relevant DSH plugins and portable capabilities before committing to an implementation.
  • Separate placeholders and prototypes from installable or tested projects.
  • Check active runtime inventory when the host safely exposes it.
  • Surface duplicates, uncertain compatibility, stale evidence, and incomplete discovery without inventing certainty.
  • Get one direct action: USE, INSTALL, COMPOSE, EXTEND, BUILD, WAIT, DISABLE, or INVESTIGATE.
  • Preserve explicit Do Not Build and Build Only boundaries for the next Agent or developer.

Other ways to use it

Agent Skill

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.

CLI

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.

Verify or remove the DSH plugin

dsh --profile web --dump-config
dsh plugin --profile web remove dsh-landscape

Stable result contract

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;
  • the existing 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.

Surfaces

SurfaceBest forEnvironment view
DSH pluginNative preflight during a Harness sessionBest-effort, read-only runtime summary
Agent SkillPreflight before an Agent changes or builds capabilityHost-dependent; safe fallback
CLINatural-language checks, automation, JSON, and briefsEcosystem-only unless supplied by a host
Node APIApplication integrationCaller-provided or unavailable
Static site/APIBrowser-safe snapshot explorationNot applicable

Node API

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.

Trust and safety

  • Discovery reads attributed public metadata; it never executes code from discovered repositories.
  • Maturity labels require observable evidence. Repository names and README claims alone do not prove a working capability.
  • Runtime inspection is best-effort and read-only. Missing host information becomes a limitation, not a setup prompt.
  • Absolute local paths, credential values, plugin configuration, and raw inspection errors are excluded from environment output.
  • Archived projects are excluded from active competition by default; forks are marked and down-ranked.

Limitations

  • DeepSeek Harness is in developer preview and can introduce breaking changes.
  • Public discovery cannot cover every GitHub, npm, private, or local project.
  • Current public runtime surfaces may not expose the active profile, DSH version, bundle provenance, or peer compatibility; Landscape reports these as unknown.
  • Metadata-based maturity is conservative and does not replace runtime acceptance testing.
  • The static website is search-only; use DSH or an Agent Skill for host-aware review.

Development

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.

License

MIT © 2026 cyanseek

评论

0
最新优先