dsh-vscode avatar

dsh-vscode

DeepSeek Harness for VS Code: right-side chat agent that reads, edits, and runs your project — like Claude Code / Codex / Copilot.

Ecosystem appDeveloperAI ModelsFileTerminal / Shell
Verification
L5 · Ecosystem app
Risk
Medium
Health
Stable
Trust
Silver

Install verification: Ecosystem app: not installed via dsh plugin add, so the L1-L5 plugin install-verification ladder does not apply; the security scan still runs.Install verification details →

Data as of (UTC)

What it doesAI

VS Code extension adding a right-side chat agent that reads, edits, and runs your project.

  • Right-side chat agent in VS Code
  • Reads, edits, and runs project files
  • Similar to Claude Code / Codex / Copilot
developerAIfileAIterminalAI

Capability tags are AI-inferred or rule-extracted from the README (see badge) — vocabulary-limited, not install-verified. AI = inferred; README = literal keyword from the readme.

Installation

GitHub repository

Ecosystem apps are not installable via dsh plugin add: desktop / web shells install or deploy through their own release channel (GitHub Releases or the project site).

The level reads L5 · Ecosystem plugin / L5 · Ecosystem app per the three-way classification (the L1-L5 ladder does not rate them); the security scan still runs.

Trends

Four lines, four questions: total stars, stars gained this week, cumulative downloads, downloads this week. The GitHub pair comes from real-read sampling — a rotating daily shard, so one plugin is re-read roughly every 5–14 days, sparser than npm; the npm pair has a value on every snapshot day. The two cumulative panels are smoothed curves with no sample markers; the two weekly panels keep their raw bars and add a 4-week moving-average trend line that strips single-week noise down to the direction of travel.

GitHub stars
4★

6 samples (roughly every 7 days), 2 of them reconstructed (dashed runs); a smoothed curve through every sample — not a daily series

09-18 → 11-14 · 6 points

Weekly star gain
+0★

Bars = the delta between two samples 5–14 days apart (negative points down, faint bars contain reconstructed points); line = 4-week moving-average trend

10-01 → 11-14 · 4 points

Cumulative downloads
—

Reconstructed backwards from the current total using daily downloads (gaps are not interpolated)

Only 0 points (< 3): no chart yet

no data yet

Weekly downloads
—

Bars = the snapshot’s last-week rolling window (official npm + npmmirror combined); line = 4-week moving-average trend — the shape of weekly demand, not a user count

Only 0 points (< 3): no chart yet

no data yet

weekly board —total board —downloads ≠ unique usersstar weekly —

Verification & Compatibility◼ Ecosystem app

Grouped by plugin version: each card lists the dsh versions actually tested against that artifact version and their install verdicts (L5 wins). Untested combinations are omitted — absence means untested, never assumed compatible. L1–L3 are a plugin-level static archive (independent of the dsh version — the same verdict on every dsh row): they sit in each row's level grid next to L4/L5, marked "static", while their evidence is stored once per plugin inside that row's "checks & evidence" instead of being duplicated per version.

External credentials (plugin-level): none detected

Install compatibility matrix (rows = plugin versions〔latest 5〕, columns = dsh versions〔latest 5〕)
◼ Ecosystem app
At-a-glance → evidence in the cards below
plugin \ dshdsh 0.1.7-rc.1dsh 0.1.5-rc.2
v0.2.0·✗
Install compatibility (by plugin version, dsh versions within)
Legend:✓ L5 runtime verified○ ecosystem plugin◼ ecosystem app✗ failed
v0.2.0github1 failed2026-11-14

https://github.com/MJ-Chang/dsh-vscode

✗dsh 0.1.5-rc.22026-09-13Ecosystem app · L5 · failed
Environment & artifact
  • • Node.js: v24.14.0
  • • DSH: 0.1.5-rc.2
  • • Plugin artifact: github · https://github.com/MJ-Chang/dsh-vscode#v0.2.0
Levels (L1–L3 plugin-level static archive · L4/L5 newest real verdicts for this combination)
L1 static
Passed
L2 static
Passed
L3 static
Passed
L4 sandbox
Passed
L5 runtime
Failed
Sandbox install failed
Ecosystem app · L5 · failed

ID: dsh-vscode@v0.2.0@dsh0.1.5-rc.2 · profile: l4-sandbox

Issued: 2026-09-13 · expires: 2026-09-20

View checks & evidence
L1 · Found✓ Passed
Defaulted from L3✓ Passed
defaulted to passed — L3 install spec passed (full-registry L3 sweep policy)
L2 · Structured✓ Passed
Defaulted from L3✓ Passed
defaulted to passed — L3 install spec passed (full-registry L3 sweep policy)
L3 · Install spec✓ Passed
Install command parsed✓ Passed
Source: registry verification · Method: install spec parse · Captured 2026-11-14
DSH version declared✓ Passed
Source: registry verification · Method: DSH version declaration · Captured 2026-11-14
L4 · Install tested✓ Passed
Install executed in confined sandbox✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-09-13
spec github:MJ-Chang/dsh-vscode#v0.2.0 · channel GitHub source · outcome ran · exit 0
check took: 28.4s
Install succeeded✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-09-13
spec github:MJ-Chang/dsh-vscode#v0.2.0 · channel GitHub source · outcome ran · exit 0
check took: 28.4s
Installed artifact confirmed✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-09-13
spec github:MJ-Chang/dsh-vscode#v0.2.0 · channel GitHub source · outcome ran · exit 0
check took: 28.4s
L5 · Run tested✕ Failed
Sandbox install passed✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-09-13
spec github:MJ-Chang/dsh-vscode#v0.2.0 · install mode reused L4 scratch
check took: 0ms
Web boot ready✕ Failed
plugin tree failed to load
HTTP endpoint served✕ Failed

L1–L3 are plugin-level static conclusions (baseline, identical across dsh versions): evidence is stored once per plugin.

Only the latest 3 published versions are shown; source-snapshot (commit hash) groups, unstated groups and older versions are omitted — the full archive stays in the data layer (snapshot/unstated groups appear only when a plugin has no published version).

Author badges

Embed the dsh.so badges in your README to show this plugin's automated security-scan and install-test status:

dsh.so risk badge
`[![dsh.so risk](https://cdn.dsh.so/badge/dsh-vscode.svg)](https://www.dsh.so/artifact/dsh-vscode/?ref=badge)`

dsh.so is an independent community project, not affiliated with or endorsed by DeepSeek AI. Badges are maintained by dsh.so; embed links carry a ?ref= attribution parameter.

Security Report

Two automated layers, neither a manual review: a static scan (per-plugin rule checks) and a model-driven dynamic security audit that covers only the subset static-flagged at warn or above and never executes plugin code.

dsh-vscodeAutomated review · daily

Static low; dynamic medium (authoritative when present) → overall medium.

static low→dynamic medium→overall medium
dynamic findings0 critical·0 high·1 medium·3 low
mediumoverall · medium
1static scan · vet primaryvet AST static analysis (npm artifact or git source)
plugin versionv0.2.0
scanned versionv0.2.0git2026-11-15
current versionv0.2.0same version
vet verdictclean · git

Vet is an AST static scan (npm artifact or git source).

show static findings (5)
  • criticalnewFunction · code-execsrc/runtime.ts:111

    sdkPromise ??= new Function('specifier', 'return import(specifier)')('@deepseek-ai/dsh-sdk-client') as Promise<SdkModule>

  • warningchildProcessImport · code-execruntime/preload-spawn.cjs:37

    const childProcess = require('node:child_process')

  • warningchildProcessImport · code-execsrc/plugins.ts:12

    import { execFile } from 'node:child_process'

  • warningchildProcessImport · code-execsrc/spawn-patch.ts:13

    const childProcess = require('node:child_process') as typeof import('node:child_process')

  • warningmanifest-fs-unfenced · sandboxpresets/minimal/agent.cordis.yml

    Mutating filesystem tool has no explicit fs backend in realm

2model reads redacted source context · never executes plugin code
highest grademedium
merged intooverall medium
audited at2026-10-17
audited artifactgit 9e6f74ad61fcbf7bf88ac43b155b45dfe28dc468
findings0 critical·0 high1 medium / 3 low hidden
show 4 audit findings
  • mediumWebview message can execute any VS Code command id without an allowlistsrc/chatView.ts:205

    The message handler passes a webview-supplied string straight to vscode.commands.executeCommand. Any script running inside the chat webview (e.g. HTML injection through unescaped assistant/markdown output, since media/chat.js could not be reviewed) can invoke arbitrary built-in or extension commands (window reload, session/workspace actions, opening external terminals, etc.). The openLink branch shows the author validates URLs by scheme, but no such validation exists here.

  • lowDynamic code construction (`new Function`) used to load the ESM SDK, but specifier is a hard-coded constantsrc/runtime.ts:111

    The construct builds a function from a string and evaluates an import; this is what static scanners typically flag as code injection. In this code the specifier is a literal and is never derived from user, workspace, or model input, so it is not exploitable — it only bypasses esbuild rewriting import() into require() for the CJS bundle. No attacker-controlled path reaches it.

  • lowPlugin install runs a user-supplied npm/git/path spec; installed package code is then executed by the runtimesrc/plugins.ts:224

    installPlugin feeds the input-box string to `npm install --no-save`, so npm lifecycle scripts and, after writeUserConfig/restart, the package's entry module run with the user's privileges inside the runtime child (sandboxed only for the agent's own tools, not for the plugin code). There is no signature, publisher or allowlist check. On Windows, the last-resort fallback spawns npm.cmd with `shell: process.platform === 'win32'` (line 67/91), and node does not escape args passed to cmd.exe, so shell metacharacters in the spec would be interpreted. Both require the user to deliberately type the sp

  • lowDebug probe RPC handlers ship in production and spawn a binary chosen by the DSH_DEBUG_NODE environment variableruntime/plugins/vscode-bridge.mjs:156

    The bridge registers debug/spawnProbe and debug/sandboxProbe (via runProbe -> subprocess.spawn(argv, ...)) that execute `SYSTEM_NODE -e <fixed probe script>`. The executable path is taken from the inherited DSH_DEBUG_NODE environment variable. Reachability is limited: only the parent extension host (owner of the child's stdio, and of the environment) can call these methods, and the extension never does, so there is no path from model output or workspace content. Risk is a shipped, undocumented process-spawning endpoint that an environment-manipulating process could steer.

One medium issue: the chat webview can invoke arbitrary VS Code command ids; plugin install, debug probes and the new Function import are user-gated and low risk.

Rule (2026-10-28): when a dynamic conclusion exists it is authoritative (it may raise or lower the level); the static scan tier (vet primary, dsh backstop) applies only when there is none (not queued / audit failed / withdrawn). The audit never writes the exemption list (exemptions are manual only); a wrong conclusion is withdrawn by human review and the tier falls back to the static scan tier.

Disclaimer: the static scan and the dynamic security audit are both automated analysis, not a security guarantee. Always review what you install.

Activity

Last commit 2026-08-21 · activity: Stable

• Repo created: 2026-08-14

• Stars: ★ 4 · Forks: 0

• Health: Stable — committed within 90 days

Source

GitHub: github.com/MJ-Chang/dsh-vscode

Frequently asked questions

What is dsh-vscode?

dsh-vscode is an ecosystem app listed on dsh.so, not a DSH plugin package: it cannot be installed with dsh plugin add and ships through the project's own release channel.

Is dsh.so or this plugin page official?

dsh.so is an independent community registry and is not affiliated with, endorsed by, or sponsored by DeepSeek AI.

How do I install dsh-vscode?

This is an ecosystem app, not a plugin installable via dsh plugin add: desktop / web shells install or deploy through their own release channel. The security scan still runs, but the L1-L5 plugin install ladder does not rate it.

How does the security scan relate to a manual audit?

Both layers are automated, not a manual audit. The static scan is heuristic analysis of the repository source, so it can produce false positives and miss context. The model-driven dynamic security audit only covers plugins static-flagged at warn or above, never executes plugin code, and its conclusion is authoritative when there is one (it may raise or lower the static tier); exemptions stay manual. Neither layer is an endorsement: review the repository and the permissions it asks for before installing, together with the verification level and health signals.

How fresh is the data on this page?

The data on this page is as of 2026-11-15 (UTC) — the most recent valid timestamp among the registry refresh, install verification, security scan and dynamic audit. Build time is never used as a data date.

Was this page helpful?