---

Compatibility

Component	Version
DeepSeek Harness	0.1.0-rc.6 (peer dependencies pinned)
Node.js	^22.19.0 || >=24.0.0
Package manager	pnpm@11.7.0
Platform	Windows / macOS / Linux (host-only plugin)
External tools	gh CLI on PATH (authenticated for API reads), npm CLI on PATH

What you get

• score tool — one target through the five-dimension pipeline; returns the structured risk card, or { kind: 'background', jobId } with background: true.
• /score command — batch scoring of a whitespace/comma-separated target list as a score-batch background job over ctx.jobs, producing a leaderboard snapshot (JSON + Markdown).
• score_report tool — fetch any stored score card (sc_...), leaderboard (lb_...), or the latest leaderboard.
• Five dimensions (weights configurable, defaults sum to 100): install success 25, maintenance 20, documentation 20, security 20, compliance 15.
• Evidence discipline — every dimension records its audit links (source, sanitized detail, observedAt); a dimension without evidence reports no-evidence (score 0, excluded from the weighted total), never a fabricated number.
• Structured results — every record carries schema: "dsh-score/v1" with first-class fields; this is the machine-readable contract downstream tooling consumes.

Quick start

Git channel

dsh plugin --profile web add github:PerryLink/dsh-score#<commit-sha>

The first add fails because pnpm blocks the package's prepare build; copy the exact key pnpm printed into the profile's pnpm-workspace.yaml and re-run:

allowBuilds:
  'dsh-score': true

npm channel

dsh plugin --profile web add dsh-score

Prebuilt packages need no build allowance. Restart the profile, then use score / /score from a session.

Install & uninstall

dsh plugin --profile web add dsh-score     # install (npm) — or the git form above
dsh plugin --profile web remove dsh-score  # uninstall

Configuration

All keys are optional (defaults shown); invalid values fail loudly at load.

Key	Default	Description
probeTimeoutMs	60000	Deadline for one gh/npm probe command.
outputTailBytes	8000	Cap on the sanitized output tail recorded per probe.
cacheMaxAgeMs	86400000	How long a cached score card is reused before re-scoring (0 disables the cache).
staleCommitWarnDays	90	Commit/publish age at which maintenance drops to warn.
staleCommitFailDays	365	Commit/publish age at which maintenance drops to fail.
staleIssueWarnDays	30	Oldest-open-issue age (response proxy) at which maintenance drops to warn.
staleIssueFailDays	180	Oldest-open-issue age at which maintenance drops to fail.
maxBatchTargets	20	/score batch cap.
batchConcurrency	1	Batch concurrency (serial avoids API-rate contention).
weights	{install:25, maintenance:20, documentation:20, security:20, compliance:15}	Per-dimension weights (each 0–100; at least one must be > 0).

Tools & surfaces

score

score(target: string, refresh?: boolean, background?: boolean)

• target — a GitHub repo (github:owner/repo, owner/repo, a git/https URL) or an npm package name.
• refresh: true bypasses the score cache and re-gathers evidence.
• background: true starts a score-batch job and returns its id.

/score <targets...>

Starts one background batch job; progress streams through the job output, and the final line names the leaderboard id for score_report.

score_report(id?)

Returns a score card (sc_...), a leaderboard (lb_...), or — with no id — the latest leaderboard.

Structured result sample

{
  "schema": "dsh-score/v1",
  "scoreId": "sc_8f1c2e4a9b3d7f01",
  "target": { "kind": "repo", "spec": "github:owner/dsh-click#abc123" },
  "scoredAt": "2026-08-16T00:00:00.000Z",
  "durationMs": 3210,
  "pluginVersion": "0.1.0",
  "dimensions": {
    "install": { "dimension": "install", "status": "no-evidence", "score": 0, "weight": 25,
                 "summary": "no dsh-test-drive result recorded for this target (install success unmeasured)",
                 "evidence": [{ "source": "test-drive", "detail": "no test-drive record found in the test_drive domain", "observedAt": "2026-08-16T00:00:00.000Z" }] },
    "maintenance": { "dimension": "maintenance", "status": "pass", "score": 100, "weight": 20,
                     "summary": "active (2026-08-10T00:00:00Z; 0 open issues)",
                     "evidence": [{ "source": "gh-api", "detail": "last activity 2026-08-10T00:00:00Z", "observedAt": "2026-08-16T00:00:00.000Z" }] }
  },
  "total": 88,
  "grade": "B",
  "verdict": "healthy (weighted total 88/100)"
}

Scoring: the total is a weighted average over dimensions that gathered evidence (no-evidence dimensions are excluded and renormalized); A ≥ 90, B ≥ 75, C ≥ 60, D ≥ 40, else F, and N/A when nothing had evidence.

Permissions & data

• Only public services are consumed: ctx.subprocess, ctx.jobs, ctx.storageDomain, ctx.tools, ctx.commands.
• Score cards and leaderboards are stored in the score storage-domain (tables scores, leaderboards; latest-leaderboard pointer). When the composition has no storageDomain (e.g. the shipped headless profile), tools still work and score persistence is disabled with a logged reason.
• Child processes inherit the provider's credential-scrubbed environment; gh reads its own credential store. No environment value is ever logged.
• All report/log strings pass through pure sanitizers: token literals, URL credentials, and bearer headers are redacted, and tails are byte-capped.

Security boundaries

• No code execution. The pipeline runs gh api and npm view only; it never installs, builds, or runs a target.
• Argv-only subprocesses. Every CLI invocation is an argv array, never shell-interpreted; repo owner/repo segments are validated against a restricted character set before use in an endpoint.
• Evidence discipline. No score is fabricated: a probe that fails or returns unparsable output yields no-evidence, never a number.
• Detection vs redaction. Secret-leak and malicious-install-script detection share the same pure regexes as redaction; both are unit-tested against extreme inputs.

Known limitations

• Repository probes require gh to be authenticated and network access to GitHub; npm probes require npm and registry access.
• A target without a resolvable GitHub repository cannot be inspected for documentation, security, or compliance (those dimensions report no-evidence).
• Install success depends on dsh-test-drive being mounted and having recorded the target; otherwise it is honestly no-evidence.
• The maintenance "issue response" signal is a proxy (oldest open issue age), not a direct response-time measurement.
• Score results are cached per target; use refresh: true (or wait past cacheMaxAgeMs) to force re-scoring.

Development

pnpm install
pnpm run typecheck && pnpm run typecheck:ci && pnpm test
pnpm run build && pnpm run verify:self-contained && pnpm run verify:artifacts && pnpm pack

• typecheck resolves @deepseek-ai/* through the local harness checkout; typecheck:ci checks against the published 0.1.0-rc.6 types.
• Tests use the real Context/Session/ToolRuntime/LocalJobRegistry/storage stack with a scripted subprocess provider.
• Real-CLI scoring (requires gh/npm on PATH, gh authenticated): invoke score from a mounted profile.
• Release: node scripts/release.mjs <x.y.z> (bumps, stamps CHANGELOG, re-runs the gate, commits + tags; never pushes).

Topics

dsh, dsh-plugin, deepseek-harness, deepseek, cordis, plugin-scoring, quality-score, leaderboard, supply-chain

Contributors

PerryLink — design and implementation.

PerryLink DSH Plugin Family

This project is one of the 29 DeepSeek Harness plugins maintained by PerryLink. If this one helps you, the others likely will too:

Plugin	One-liner
dsh-auto-review	Second-model auto-review on the approval chain, fail-closed by default
dsh-background-agents	Durable background child agents with a Web UI sidebar, messaging and interrupt
dsh-budget	Cost governance for DeepSeek Harness: budgets, carbon, and latency in one panel.
dsh-checkpoint-rewind	Claude Code /rewind-equivalent: snapshots, session forks, one-shot restore
dsh-claude-move	Migrate Claude Code sessions, memory, skills and CLAUDE.md into DSH
dsh-click	Cross-platform native desktop control for DeepSeek Harness — Windows first.
dsh-composer-history	Terminal-style input history for the web composer: arrows, Ctrl+R search
dsh-defend	Prompt-injection, jailbreak, and secret-leak defense for DeepSeek Harness.
dsh-doublecheck	Engineering-discipline guard: requirements grill, test gates, adversary review
dsh-draw	Unified static-image generation routing for DeepSeek Harness.
dsh-fast	Read-only performance diagnostics for DeepSeek Harness.
dsh-github	GitHub PR/issues integration for DSH, every write gated by approval
dsh-library	Local document knowledge base for DeepSeek Harness.
dsh-local-ai	Local-model (Ollama) integration for DeepSeek Harness.
dsh-lsp-actions	LSP diagnostics, formatting, completion, code actions and rename over language servers
dsh-mask	PII masking middleware for DeepSeek Harness — anonymize personal data before it reaches the model, restore it at the display layer.
dsh-mcp-panel	Read-only MCP runtime panel: /mcp command + Settings tab with status, tools and errors
dsh-memento	Approval-gated cross-session memory: ctx.memory seam + SQLite + memory tool
dsh-observe	OpenTelemetry and Langfuse observability exporter for DeepSeek Harness.
dsh-output-styles	Claude Code outputStyles-equivalent runtime style switching
dsh-permission-rules	Claude Code-style declarative allow/deny/ask permission rules with audit
dsh-plugin-guide	Plugin-development knowledge base as an on-demand agent skill
dsh-score	Multi-dimensional quality scoring for DeepSeek Harness plugins.
dsh-session-pin	Pin sessions in the Web sidebar with durable ordering
dsh-session-sync	Cross-device session sync for DeepSeek Harness — a dedicated git mirror of your session store.
dsh-skill-pack-security	Security-audit skill pack: secret scan, dependency and supply-chain review
dsh-talk	Voice-first session loop for DeepSeek Harness: talk to it, hear it answer.
dsh-test-drive	Isolated install-and-smoke test drives for DeepSeek Harness plugins.
dsh-translate	Vendor parameter translation and deterministic JSON repair for DeepSeek Harness.

License

Apache-2.0