@121212165/dsh-plugin-subagent-roster

Constrained subagent roster for DeepSeek Harness: one tool per role, each pinning a child system prompt, a capability allow-list, and a model tier. Bulk work vs team work by tool shape. 子代理名册：按角色约束系统提示词/能力范围/模型档位。

PluginAI ModelsAgent orchestration

Verification

L5 · Run tested

Verified

Risk

LowRisk

Health

Active

Trust

Gold

Install verification: runtime verified (L5) · tested on dsh 0.2.1-rc.2 · last checked 10-21Install verification details →

What it does

Installation

dsh plugin --profile web add github:121212165/dsh-plugin-subagent-roster#v0.2.2

verified v0.2.2

Install method: GitHub · runtime verified (L5)

Verification & Compatibility

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〕)

At-a-glance → evidence in the cards below

plugin ＼ dsh	dsh 0.2.1-rc.2 current
v0.2.2	✓

Install compatibility (by plugin version, dsh versions within)

Legend:✓ L5 runtime verified○ ecosystem plugin◼ ecosystem app✗ failed

v0.2.2github✓ 1 runtime-verified2026-10-21

https://github.com/121212165/dsh-plugin-subagent-roster

✓dsh 0.2.1-rc.2current2026-10-21L5 · runtime verified

Environment & artifact

• Node.js: v24.14.0
• DSH: 0.2.1-rc.2
• Plugin artifact: github · https://github.com/121212165/dsh-plugin-subagent-roster#v0.2.2

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

Passed

Install + runtime smoke verified in the dsh sandbox — installs and runs

L5 · runtime verified · valid for dsh 0.2.1-rc.2

ID: dsh-plugin-subagent-roster@v0.2.2@dsh0.2.1-rc.2 · profile: l3-fullscan

Issued: 2026-10-21 · expires: 2026-10-28

View checks & evidence

L1 · Found✓ Passed

Defaulted from L3L1.0_DEFAULTED_FROM_L3✓ Passed

defaulted to passed — L3 install spec passed (full-registry L3 sweep policy)

L2 · Structured✓ Passed

Defaulted from L3L2.0_DEFAULTED_FROM_L3✓ Passed

defaulted to passed — L3 install spec passed (full-registry L3 sweep policy)

L3 · Install spec✓ Passed

Install command parsedL3.1_INSTALL_COMMAND_PARSED✓ Passed

Source: registry verification · Method: install spec parse · Captured 2026-10-21

DSH version declaredL3.2_DSH_VERSION_DECLARED✓ Passed

Source: registry verification · Method: DSH version declaration · Captured 2026-10-21

L4 · Install tested✓ Passed

Install executed in confined sandboxL4.1_INSTALL_EXECUTED_CONFINED✓ Passed

Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-19

spec github:121212165/dsh-plugin-subagent-roster#v0.2.2 · channel GitHub source · outcome ran · exit 0

check took: 2.2s

Install succeededL4.2_INSTALL_SUCCESS✓ Passed

Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-19

spec github:121212165/dsh-plugin-subagent-roster#v0.2.2 · channel GitHub source · outcome ran · exit 0

check took: 2.2s

Installed artifact confirmedL4.3_ARTIFACT_CONFIRMED✓ Passed

Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-19

spec github:121212165/dsh-plugin-subagent-roster#v0.2.2 · channel GitHub source · outcome ran · exit 0

check took: 2.2s

L5 · Run tested✓ Passed

active in loader tree (150 entries)

Sandbox install passedL5.1_INSTALL_PASSED_SANDBOXED✓ Passed

Source: Windows ACL sandbox · Method: sandbox install run · Captured 2026-10-19

spec github:121212165/dsh-plugin-subagent-roster#v0.2.2 · install mode reused L4 scratch

check took: 114ms

Web boot readyL5.2_WEB_BOOT_READY✓ Passed

Source: DSH runtime · Method: web boot · Captured 2026-10-19

ready line dsh web: http://127.0.0.1:57829/?token=…

HTTP endpoint servedL5.3_HTTP_SERVED✓ Passed

Source: DSH runtime · Method: HTTP request · Captured 2026-10-19

path / · HTTP 200 · took 114ms

Plugin active in inventoryL5.4_PLUGIN_INVENTORY_ACTIVE✓ Passed

Source: DSH runtime · Method: plugin inventory query · Captured 2026-10-19

verdict active · entries 150

match: include:subagent-roster · dsh-plugin-subagent-roster · phase active

phases: active 122 · null 28

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

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-plugin-subagent-rosterAutomated 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 versionv0.2.2dsh manifest

scanned versionv0.2.2git2026-10-21

current versionv0.2.2same version

vet verdictclean · git

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

show static findings (11)

warningnetIp · networkbench/gateway.mjs:84
const url = new URL(req.url, `http://127.0.0.1:${PORT}`)
warningnetIp · networkbench/gateway.mjs:277
console.log(`[gateway] listening on http://127.0.0.1:${PORT} keys=${keys.length} budget=${BUDGET} upstream=${UPSTREAM}`)
warningfsWrite · filesbench/gateway.mjs:59
fs.appendFileSync(LOG, JSON.stringify(entry) + '\n')
warningfsWrite · filesbench/regrade.mjs:45
fs.writeFileSync(target, `${JSON.stringify(out, null, 2)}\n`)
warningchildProcess · code-execbench/run.mjs:121
const result = spawnSync(
warningfsWrite · filesbench/run.mjs:71
fs.writeFileSync(target, body)
warningfsWrite · filesbench/run.mjs:235
fs.writeFileSync(outFile, JSON.stringify(summary, null, 2))
warningchildProcess · code-execbin/roster-fanout.mjs:66
const child = spawn(process.execPath, [DSH_BIN, '--profile', profile, '--json', taskFor({ role, item, contract, index, total: items.length }
warningfsWrite · filesbin/roster-fanout.mjs:116
fs.writeFileSync(path.resolve(outFile), results.map((result) => JSON.stringify(result)).join('\n') + '\n')
warningfsWrite · fileslib/store.js:106
fs.writeFileSync(file, `${JSON.stringify(ROLES, null, 2)}\n`, 'utf8')
warningfsWrite · fileslib/store.js:134
fs.writeFileSync(temporary, `${JSON.stringify(document, null, 2)}\n`, 'utf8')

2dynamic auditmodel reads redacted source context · never executes plugin code

highest grademedium

merged intooverall medium

audited at2026-10-19

findings0 critical·0 high2 medium / 1 low hidden

show 3 audit findings

mediumUnauthenticated loopback roster-editing HTTP page runs by default and its handler (lib/page.js) is not in the audited sourcelib/index.js:62
The page persists an arbitrary roster. Saved role `prompt` becomes the child's persona and saved `description`/`id` are rendered into the parent's system prompt section by routingText(), so anyone who can reach 127.0.0.1:4317 (another local process, or a browser page issuing a cross-site request if page.js lacks Origin/Content-Type checks — page.js was not provided for review) can persist instructions into every subsequent agent turn and change model routes and tool filters within the CAPABILITIES allow-list.
mediumdelegate resolves roles from the shipped ROLES map, so capability/tier changes written to roles.json are silently not enforcedlib/delegate.js:18
registerDelegate builds the `role` enum from the loaded/edited roster, but execute() looks the role up in the module-level map built from the built-in ROLES and produces persona/toolFilter/agentOptions from that. An operator who removes a capability (e.g. write) from a role, or changes its tier, through the documented roster file or page gets a successful save and an unchanged child: the restriction is silently absent while the UI and routing text claim it applies. The reverse case throws 'unknown role' for newly added roles whose id is not shipped.
lowDev bench gateway is an unauthenticated loopback proxy that spends stored upstream API keys and parses any bodybench/gateway.mjs:80
The server accepts POST */chat/completions with no auth, origin, or content-type check, round-robins the OpenRouter keys read from OR_KEYS_FILE and writes one JSONL line per call (including the first 4000 chars of each system prompt) to bench/state/calls.jsonl. Any local process, or a browser page making a simple cross-site POST, can consume the account's quota and cause prompt content to be persisted to disk. Scope is limited: bench/ is excluded from the published package files and the listener is bound to 127.0.0.1.

Delegation enforces the built-in ROLES instead of the operator-edited roster, and the default loopback roster-editing HTTP page is not included in the audited source.

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 2026-10-18 · activity: Active

• Repo created: 2026-10-18

• Stars: ★ 1 · Forks: 0

• Health: Active — committed within last 30 days

Source

GitHub: github.com/121212165/dsh-plugin-subagent-roster

Plugin notes

What is the @121212165/dsh-plugin-subagent-roster DeepSeek Harness plugin?

@121212165/dsh-plugin-subagent-roster is a DeepSeek Harness plugin maintained by @121212165. It is cataloged on dsh.so as a AI Models plugin for Agent orchestration tasks, with a public GitHub source, declared install command, and automated static scan results; plugins static-flagged at warn or above also carry a model-driven dynamic security audit record.

How do I install @121212165/dsh-plugin-subagent-roster?

Run dsh plugin --profile web add github:121212165/dsh-plugin-subagent-roster#v0.2.2 in a DeepSeek Harness (dsh) environment. This exact command has been verified by a real install inside an isolated windows-acl sandbox (L4) plus web smoke testing (L5).

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 (it can produce false positives and miss context), and the dynamic security audit has a model read the harvested source context with secret files stripped and content redacted inline — plugin code is never executed. The dynamic audit only covers plugins static-flagged at warn or above, and its conclusion can only raise the combined verdict, never exempt it. Neither layer is a manual security audit or an endorsement; before using a plugin, review the repository and installed permissions, and combine both with the verification level (L4/L5) and health signals.

"DeepSeek Harness (dsh) is an open-source agent harness developed by DeepSeek AI. It uses an architecture where everything is a plugin, and is powered by Cordis."
— DeepSeek Harness README

Sources: GitHub repository · DeepSeek Harness README · security scan data · Maintained by dsh.so · Last updated 2026-10-18

@121212165/dsh-plugin-subagent-roster

What it does

Installation

Verification & Compatibility

Author badges

Security Report

Activity

Source

Author README