ang-XWBWZ/dsh-approval-ai
AI approval answerer for DeepSeek Harness (DSH) using the unified LLM route with fail-closed policy checks.
AI approval answerer for DSH's approval/request flow. DSH creates and routes the approval request; this plugin only answers that already-created request with a minimal ctx.llm assessment — it adds no path, workspace, shell, sandbox, or risk trigger of its own. Modes: ai-only (valid allow assessments map to allowed-once; deny, high/critical risk, review, malformed output, missing routes, cancellation, timeout, and provider errors fail closed), ai-then-human (review maps to a configured outcome), and human-only (delegates every request). Config is supplied, never read from files. Also registers the native /approval-ai slash command for live changes.
Install
dsh plugin --profile web add @llangtop/dsh-approval-ai@nextnpm package @llangtop/dsh-approval-ai — 0.1.0-rc.5 verified on registry (2026-08-27); the README states 0.1.0-rc.6 is a prerelease and should use the next dist-tag: dsh plugin --profile web add @llangtop/dsh-approval-ai@next. Install 0.1.0-rc.6 or later (0.1.0-rc.5 has no client card) and restart Web. The source repo is github.com/ang-XWBWZ/dsh-approval-ai.
Compatibility
DeepSeek Harness web profile, 0.1.0-rc.6 or later. Default route: provider deepseek-official, model deepseek-v4-flash.
Details
- Repo: ang-XWBWZ/dsh-approval-ai
- Category: Agent Capabilities
- Stars: 3
- Version: npm @llangtop/dsh-approval-ai 0.1.0-rc.6 (next dist-tag; rc.5 registry-verified 2026-08-27)
- Last push: 2026-08-14
- First seen: 2026-08-14
Recent updates
AI answers approval/request via ctx.llm; ai-only / ai-then-human / human-only modes; fail-closed on any uncertainty; /approval-ai command.
FAQ
- Does it add risk triggers?
- No — it adds no path, workspace, shell, sandbox, or risk trigger of its own; it only answers approval requests DSH already created.
- What happens on uncertainty?
- Fail-closed: deny, high/critical risk, review, malformed output, missing routes, cancellation, timeout, and provider errors all deny.
- Which mode should I use?
- ai-only auto-approves low-risk requests; ai-then-human routes 'review' to a human outcome; human-only delegates everything to a human.
Alternatives
Andy8647/dsh-auto-approval · timeance/dsh-approve-for-me · pandashere/dsh-self-control-guard