Leeaoyin/dsh-agent-harness-audit ↗★ 0
dsh-harness-audit
Audit an agent harness against the harness-evaluation criteria, with machine-enforced evidence validation.
AI 분석
核心用途是根据评估标准对 Agent 进行合规性审计。适合需要对 Agent 进行自动化检查、并发控制和生成审计报告的 DSH 开发者。
설치
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:Leeaoyin/dsh-agent-harness-auditConfiguration
| Field | Default | Meaning |
|---|---|---|
checks | [] | Check ids to run. Empty means "ask", never "audit everything". |
priorityFloor | 2 | Run checks at or below this priority (1 = P1 only). An explicitly named check runs regardless. |
concurrency | 3 | Checks in flight at once. 1 is supported and is the cheapest way to debug. |
subagentProvider | spawn | Subagent provider name. |
maxTokensPerCheck | 120000 | Advisory. See the limitations below. |
outputDir | .harness-audit | Workspace-relative report directory. |
announceOnStart | true | Post a one-line notice when a run starts, so a fresh session has something to attach the job UI to. |
background | true | Run as a job. false blocks the command until the audit finishes. |
useLsp | true | Try LSP navigation, degrading to text search. |
crossCheckAnalysis | false | Reserved; not yet emitted. |
language | auto | Output language. auto follows the harness locale, then the host locale, then English. |
excludePaths | dependency dirs | Directory names refused as out of scope. [] audits a vendored framework deliberately. |