Plugins

Second-model auto-review on the approval answerer chain: a read-only reviewer subagent returns structured allow/deny verdicts with reasons, fail-closed by default.

Multi-agent workflow for math-modeling contests: the "UltraMath" preset runs a problem from reading through framework, derivation, coding, verification, paper and review, plus 4 single-stage role presets (mathematician, engineer, writer, reviewer) and 33 model-library skill docs.

Rule-gated automatic approval for DeepSeek Harness sandbox escalations with an optional LLM reviewer and native human fallback.

Template-driven code review skill: five-step review flow, critical/medium/minor severity grading, four-dimension scoring (quality, security, performance, maintainability), dual text and HTML report output, and an on-demand reference knowledge base.

Role-based multi-agent team for DeepSeek Harness. The /team command assembles up to 12 specialist subagents (product, architect, researcher, UI, backend, frontend, database, security, reviewer, QA, DevOps, docs) and runs a nine-phase gated pipeline (clarify, research, design, spec review, plan approval, implement, review, test, deliver) over a shared workspace of artifacts (SPEC, PLAN, TASKS, STATE, REVIEW, TEST). Quality gates are enforced by plugin code rather than requested in prompts: an unfinished task cannot be marked completed, quality findings must be adjudicated by the reviewer or QA role, and coverage gaps or rework over budget appear live in the overlay and in /team status. Ships a live team overlay (phases, roster with per-role models, task DAG, artifact preview, decision buttons), a persistent-team mode that resumes across sessions, and per-run token and time accounting. Zero runtime dependencies, no build step.

DeepSeek Harness bundle registering one skill that turns long-form production (academic papers, industry analysis, business commentary) into a 9-role pipeline: T1 literature / T2 data / T3 case scouts running in parallel, T4 analyst, T5 writer, T6 critical companion, T7 auditor, T8 finalizer run by the coordinator, T9 peer reviewer — across 6 phases with triangular evidence verification, a 23-check M-gate, a G0-G14 audit including a Chinese AI-trace gate, peer-review scoring with journal matching, and 4 human checkpoints. Installs with `dsh plugin add lunheng-article-pipeline` (package.json#dsh.bundle.patch plus a lib/index.js entry that registers the skill); ships role cards for T1-T9, templates and 11 zero-dependency verification scripts, with five-language READMEs.

Model-based permission approval: an approval-request answerer backed by a separate reviewer model.

Role-based AI swarms for DeepSeek Harness: an architect plans, parallel builders execute, reviewers gate quality, and an integrator ships. Per-role model pinning with fallback chains, review loops, evidence contracts, human-gated reviews, quota-aware pause/resume, and a live kanban + flow chart in the web GUI. 40 tests; installs from GitHub or npm.

A governed swarm of six specialist DSH agents — orchestrator, planner, knowledge-base bridge, developer and two reviewers — turns a requirement into a strict phase pipeline where every delivery must prove itself with structured evidence before the next phase may start; reviewers are read-only, code merges only after a review pass, and an append-only event log gives full replay and audit on a live kanban board. Its design draws inspiration from the Hermes Agent kanban.

Runs work as a crew of role agents: a product manager writes the PRD or DoD, waits for your confirmation, then starts architect, engineer, QA and reviewers whose tool sets are locked per role, and shares work through files on disk.

Adds an Auto Approve permission preset to the Web UI, using a fresh restricted Reviewer Agent to allow or deny each approval request.

Agentic answer reviewer for DeepSeek Harness: each final assistant turn is re-reviewed by a separate LLM graded 1-100; below-threshold scores steer the agent back with concrete feedback. The score is shown as a chip on the answer's action row (green pass / red needs-work, with retry count and reason on hover). Live gate config over a 127.0.0.1 HTTP server, plus a collapsed dock above the composer and an optional right-sidebar tab via dsh-better-sidebar.

Pairs the running executor with a stronger reviewer model through zero-parameter advisor() calls and step-based patrol checks, returning plan, correction, or stop guidance.

GPT plans dependency DAGs, DeepSeek workers execute tasks with adaptive parallelism, and an optional GPT reviewer audits incremental Git diffs and drives repair rounds.

Unofficial FreeCodeGo for DeepSeek Harness: a managed free-model gateway and provider catalogs, image and video generation behind one settings switch, DeepSeek/Codex/Claude engine routing, an independent advisor reviewer, code review, code graphs, project memory, and an engineering team toolchain.

Audit each finished assistant turn with an independent reviewer model and steer the agent to fix failed output, with at most 2 review rounds per turn and SOP-folder standards injected as an extra review dimension.

Filesystem permissions for shell calls, granted per path and capability rather than per command name. The command line is parsed for the read, write, create, delete and execute effects it needs, a capability it lacks raises the approval card instead of a flat refusal, and the same rules are compiled into the macOS Seatbelt profile the process, its children and the code its arguments never showed all run under. An Approvals tab in the conversation reads the audit log back and shows which rule, automatic reviewer or person answered each call.

Mandatory cross-review for design proposals and lessons-learned: auto-detects design doc writes and enqueues them (tier=review) into the task queue, delivers the external reviewer verdict back to the initiating session, and appends approved preventive measures to OPS-GUARDRAILS.md (append-only, conflict-checked).

Runs a parallel code review squad of read-only reviewer subagents (security, correctness, tests, style), each optionally on its own model, and aggregates one severity-grouped report from /review or a code_review tool call.

LLM approval answerer for sandbox escalations beyond workspace-write: a deterministic regex filter first, then a clean-context reviewer model; humans are asked only when it is unsure. Requires the bundled core patches.

Independent Codex Guardian-style approval reviewer for DSH.

Fail-closed LLM-assisted approval reviewer for DeepSeek Harness

Codex-style auto-review permission mode: adds an auto-review preset that auto-approves safe sandbox escalations, asks on risky or ambiguous ones, and rejects critical unconfirmed operations.

Headless Claude Code consultations with advisor, reviewer, and designer roles, returned as reference answers.