Use the archify skill to map this repository's runtime architecture.
Show 8–12 core components, one primary path, external dependencies, and trust boundaries.
Put supporting detail in cards instead of adding more edges.
After delivery, return the exact workspace paths of the specification JSON and the HTML artifact.
卸载:
dsh plugin --profile remove dsh-archify
Produced Files 限制: shell 命令生成的文件不会自动出现在 Web 端的 Produced Files 栏,
请让 agent 返回规格 JSON 与 HTML 工件的精确工作区路径,再从工作区打开(上游官方 DSH 包同样如此)。
One line: a diagramming agent skill for DeepSeek Harness — turn a small typed JSON spec into a validated architecture, workflow, sequence, data-flow, or lifecycle diagram as a self-contained HTML artifact.
A port of tt-a1i/archify (MIT, v2.14.0; same source as the official @tt-a1i/archify-dsh@0.1.0).
The skill payload (archify/ and examples/) is byte-identical to upstream; only the dsh packaging and naming are adapted.
✨ Features
🗺️ Five diagram types — architecture / workflow / sequence / dataflow / lifecycle from one JSON spec
✅ Verifiable — deterministic validate with 9-check showcase or basic acceptance receipts
The prepare script stages archify/ into skills/ on install. If pnpm blocks the build step,
add this package to allowBuilds in the profile's pnpm-workspace.yaml and re-run.
From a local directory (run npm install first to trigger staging):
Requires Node.js ^22.19.0 || >=24.0.0 and dsh 0.1.0-rc.6.
🚀 Usage
Ask the agent to load the skill by name:
Use the archify skill to map this repository's runtime architecture.
Show 8–12 core components, one primary path, external dependencies, and trust boundaries.
Put supporting detail in cards instead of adding more edges.
After delivery, return the exact workspace paths of the specification JSON and the HTML artifact.
Uninstall:
dsh plugin --profile remove dsh-archify
Produced Files limitation: files created by shell commands do not automatically appear in
the Web Produced Files strip — ask the agent to return the exact workspace paths of the
specification JSON and the HTML artifact, then open them from the workspace (same as the
official upstream DSH package).
Security posture: no telemetry, network client, credentials handling, or background
service; no install / postinstall / preinstall lifecycle hooks (only prepare stages
the skill files).
Port notes (vs upstream tt-a1i/archify @ archify-dsh-v0.1.0)
Byte-identical, diff-verifiable: archify/ (SKILL.md, renderers, schemas, bin, examples,
recipes, references, assets, delta, scripts, test) and the root examples/ are upstream files.
Verify it yourself — don't take our word — THIRD_PARTY_NOTICES.md
pins the upstream release tag, commit, and per-file SHA-256 with copy-paste compare commands;
test/zero-regression.test.mjs re-checks the pins on every npm test.
Adaptations (official @tt-a1i/archify-dsh@0.1.0 → dsh-archify): package name and resolution
path; added prepare (upstream ships a prebuilt tarball; a git install must stage skills/
itself); upstream repo-root website/release files (docs/, README_EN.md, README_ZH.md,
benchmarks/, .github/, …) are not part of the skill payload and were not ported.
Known limitation: skill copy is English — that is upstream's original text, kept verbatim.
🛠 Development
npm install # prepare stages skills/archify
npm test # adapter tests + skill's own tests (golden + self-contained units)
npm run test:skill # skill tests only
node scripts/pack.mjs --out /tmp/dsh-archify.tgz # build a tarball locally
scripts/run-skill-tests.mjs skips 19 skill tests that cannot run in the port repo: 18
bound to upstream repo-root website/release files (docs/, README_ZH.md, benchmarks/,
ROADMAP.md, .github/, …; list inside the file) plus generate-validators.test.mjs,
which needs the ajv devDependency the adapter deliberately does not carry (upstream
security posture; the committed generated-validators.mjs is hash-pinned in
THIRD_PARTY_NOTICES.md and re-verified on every npm test). The skipped test files
remain untouched under archify/test/.
📄 License
MIT. Upstream tt-a1i/archifyCopyright (c) 2026 tt-a1i (Archify), Copyright (c) 2025 Cocoon AI,
this port Copyright (c) 2026 GongYuanCaiJi (dsh port). See LICENSE.