dsh.so
EN
English中文
@lemonmmice/dsh-agent-toolchain avatar

@lemonmmice/dsh-agent-toolchain

Evidence over claims for coding agents: 55 DSH/MCP tools (23 read-only) that build, drive, capture and profile a live Windows desktop app, then adjudicate what the agent claims against real evidence. Failure corpus included.

Ecosystem pluginDeveloperDesktop
Verification
L5 · Ecosystem plugin
Ecosystem plugin
Risk
LowRisk
Health
Active
Trust
Silver

Install verification: Ecosystem plugin: installs via dsh plugin add but declares no dsh.bundle and mounts no plugin capability — the L1-L5 install ladder does not rate it; the security scan still runs.Install verification details →

What it does

Evidence over claims for coding agents: 55 DSH/MCP tools (23 read-only) that build, drive, capture and profile a live Windows desktop app, then adjudicate what the agent claims against real evidence. Failure corpus included.

Installation

dsh plugin --profile web add github:lemonmmice/dsh-agent-toolchain#v0.1.0
verified v0.1.0

Install method: GitHub · runtime test inconclusive (L5 unknown)

Verification & Compatibility○ Ecosystem plugin

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

Declared DSH compatibilityfrom package.json · peerDependencies
author declared no dsh version requirement

○ The author declares no dsh version requirement — install is not blocked by version.

Install compatibility matrix (rows = plugin versions〔latest 5〕, columns = dsh versions〔latest 5〕)
○ Ecosystem plugin
At-a-glance → evidence in the cards below
plugin ＼ dshdsh 0.2.1-rc.2
current
v0.1.0○
Install compatibility (by plugin version, dsh versions within)
Legend:✓ L5 runtime verified○ ecosystem plugin◼ ecosystem app✗ failed
v0.1.0github○ 1 ecosystem plugin2026-10-20

https://github.com/lemonmmice/dsh-agent-toolchain

○dsh 0.2.1-rc.2current2026-10-20Ecosystem plugin
Environment & artifact
  • • Node.js: v24.14.0
  • • DSH: 0.2.1-rc.2
  • • Plugin artifact: github · https://github.com/lemonmmice/dsh-agent-toolchain#v0.1.0
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
Unknown
Ecosystem plugin: no install verdict — not a usable dsh plugin (shown as L5 · ecosystem plugin)
Ecosystem plugin

ID: dsh-agent-toolchain@v0.1.0@dsh0.2.1-rc.2 · profile: l4-sandbox

Issued: 2026-10-20 · expires: 2026-10-27

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 2026-10-18
DSH version declared✓ Passed
Source: registry verification · Method: DSH version declaration · Captured 2026-10-18
L4 · Install tested✓ Passed
Install executed in confined sandbox✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-20
spec github:lemonmmice/dsh-agent-toolchain#v0.1.0 · channel GitHub source · outcome ran · exit 0
check took: 1.7s
Install succeeded✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-20
spec github:lemonmmice/dsh-agent-toolchain#v0.1.0 · channel GitHub source · outcome ran · exit 0
check took: 1.7s
Installed artifact confirmed✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-20
spec github:lemonmmice/dsh-agent-toolchain#v0.1.0 · channel GitHub source · outcome ran · exit 0
check took: 1.7s
L5 · Run tested? Unknown

not present in web-profile loader inventory

Sandbox install passed✓ Passed
Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-20
spec github:lemonmmice/dsh-agent-toolchain#v0.1.0 · install mode reused L4 scratch
check took: 22ms
Web boot ready✓ Passed
Source: DSH runtime · Method: web boot · Captured 2026-10-20
ready line dsh web: http://127.0.0.1:64444/?token=…
HTTP endpoint served✓ Passed
Source: DSH runtime · Method: HTTP request · Captured 2026-10-20
path / · HTTP 200 · took 22ms
Plugin active in inventory? Unknown
Source: DSH runtime · Method: plugin inventory query · Captured 2026-10-20
verdict not-found · entries 149
phases: active 121 · null 28
not present in web-profile loader inventory

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

Author badges

Embed the official badges in your README to showcase security & install-test status:

dsh.so risk
`[![dsh.so risk](https://www.dsh.so/badge/dsh-agent-toolchain.svg)](https://www.dsh.so/artifact/dsh-agent-toolchain/)`

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-agent-toolchainAutomated review · daily

Static low, dynamic medium — stricter wins → overall medium.

static low→dynamic medium→overall medium
0 critical·0 high
mediumoverall · medium
1static scan · vet primaryvet AST static analysis (npm artifact or git source)
Version binding
plugin versionv0.1.0
scanned versionv0.1.0git2026-10-21
current versionv0.1.0same version
vet verdictclean · git

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

show static findings (4)
  • warningchildProcess · code-execmcp/smoke.mjs:9

    const child = spawn(process.execPath, [join(here, 'server.mjs')], { stdio: ['pipe', 'pipe', 'pipe'] })

  • warningfsWrite · filesplugins/dsh-memory/lib/embed-provider.mjs:69

    fs.writeFileSync(path.join(this.cacheDir, "emb-" + (h >>> 0).toString(36) + ".json"), JSON.stringify({ vector }));

  • warningfsWrite · filesplugins/dsh-memory/lib/kv.mjs:17

    _write(rows) { fs.writeFileSync(this.file, rows.map(r => JSON.stringify(r)).join("\n") + "\n"); }

  • warningfsWrite · filesplugins/dsh-memory/lib/store.mjs:21

    fs.writeFileSync(this.file, rows.map(r => JSON.stringify(r)).join("\n") + "\n");

2dynamic auditmodel reads redacted source context · never executes plugin code
highest grademedium
merged intooverall medium
audited at2026-10-19
findings0 critical·0 high3 medium / 1 low hidden
show 4 audit findings
  • mediummemory_index accepts any absolute path, letting the agent ingest credential/settings filesplugins/dsh-memory/index.js:48

    indexWorkspace (plugins/dsh-memory/lib/memory.mjs:37-49) walks whatever root it is given with no confinement to a project, and the extension filter in `else if (/\.(md|txt|json|yaml|yml|cs|ts|js|mjs|py|xaml|xml|html|vue|sql)$/i.test(e.name))` admits .yaml/.yml/.json/.txt, so ~/.dsh/.credentials.yaml and ~/.dsh/settings.yaml are readable. Chunk text is then embedded (forwarded to MiniMax when a key exists), persisted under ~/.dsh/memory, and echoed back to the model by memory_search (400 chars per chunk). Repo content that the model is asked to index can therefore steer it into sweeping a home

  • mediumPrompt claims data never leaves the machine, but indexed text is POSTed to api.minimax.chatplugins/dsh-memory/lib/embed-provider.mjs:60

    When MINIMAX_CN_API_KEY is present (env or ~/.dsh/.credentials.yaml), every chunk of indexed file content is sent to https://api.minimax.chat/v1/embeddings (line 57). The systemPrompt registered in plugins/dsh-memory/index.js:20 states '数据全部存储在本地 ~/.dsh/memory/，不外传。', so both the user and the model are told the opposite of the actual data flow when deciding whether to index private code.

  • mediumhttp_request core has no host allowlist and follows redirects, enabling SSRF from the agentplugins/dsh-postman/lib/http.mjs:66

    The only validation is `if (parsed.protocol !== 'http:' && parsed.protocol !== 'https:')` (line 56); there is no loopback/LAN/metadata host filtering and the request uses `redirect: 'follow'` (line 71), so a URL that looks public can redirect to 127.0.0.1, 169.254.169.254 or an intranet service, and up to 2 MiB of the response body is returned into the agent context. The postman/ui-drive index.js wrapper and mcp/server.mjs were not harvested, but the package description and cordis patch state http_request is exposed as an agent tool over this core.

  • lowMemory stores are plaintext with umask-default permissions and parsed without validationplugins/dsh-memory/lib/kv.mjs:14

    KV memories and indexed chunks (which may include secrets pulled in via the memory_index path issue) are written to ~/.dsh/memory with default file modes and no encryption; an unguarded JSON.parse on each line means a corrupted or externally modified kv.jsonl / vector JSONL makes every memory tool throw instead of degrading.

Agent tools can index arbitrary local paths (incl. credentials) whose text is sent to MiniMax despite a 'stays local' prompt claim; http_request allows unconstrained SSRF.

Rule: the dynamic audit can only raise risk, never lower it (raise-only, never an exemption); a wrong raise is withdrawn by human review and the tier falls back to the static verdict. Overall verdict = the stricter of the two layers.

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 2026-10-19 · activity: Active

• Repo created: 2026-09-16

• Stars: ★ 3 · Forks: 0

• Health: Active — committed within last 30 days

Source

GitHub: github.com/lemonmmice/dsh-agent-toolchain

Was this page helpful?