플러그인

RLM mode for dsh: recursive sub-agents via a unified rlm() function — spawn, await, and chain child agents as first-class Python calls. Adds a persistent IPython kernel with tools.* bindings and context-as-variable.

Verification toolkit for DSH agents: evidence-based claim checking against workspace files with line citations, config validation (JSON/YAML), and read-only URL/npm/GitHub submission-readiness probes.

Adversarial code review — parallel finders inspect a diff through separate lenses (correctness, lifecycle, contract, security), then each finding goes to independent verifiers tasked with refuting it, and one refutation drops it.

GitHub REST API tools + browser sidebar panel: repos, search, issues, PRs, comments, account binding (credential-store), and one-click clone-as-workspace.

Collects primary frontier-AI sources and gates reproduction with evidence, rubrics, rollback, and manifests.

Declarative task-DAG orchestration for DSH: validates dependency graphs, runs topological task layers in parallel through workflow-backed subagents, and propagates failures deterministically.

Chain-of-thought leakage linter: zero-dependency CLI that finds AI-session residue in docs and comments (dead design citations, PR vantage, change narration, review choreography), shipping the cot-trim fixing skill as the installable bundle.

Independent browser acceptance testing for agent deliverables: JSON spec in, real-Chromium verdict out.

Human-in-the-loop tools (`human_task` / `human_task_ready_check`): pause the agent until the user performs a real-world action or returns an observation (GUI operation, game testing, visual verification, hardware checks), gated by session consent and an AFK presence check, returning a structured JSON result via a web task dialog.

Reuse-first capability evolution for agents: resolve existing tools and skills, discover and review exact-commit DSH plugins, install with one-time approval, verify real tool round trips, and minimally adapt partial matches instead of rebuilding from scratch.

Paddock: constrained local filesystem MCP server for agents — text read/write, directory ops, filename and content search confined to configurable zones, with zero runtime dependencies and a dsh bundle.

Doctrove: versioned library documentation retrieval MCP server for coding agents — catalog lookup, version selection and focus-ranked doc extraction, with zero runtime dependencies and a dsh bundle.

Translation toolkit for DSH agents: sentence segmentation, term and glossary extraction, source-target QA (numbers, units, brackets), tone guides, and a durable workspace translation memory.

Local-life toolkit for DSH agents: budget planning, bill splitting, unit-price and discount math, common unit conversions (jin/li/chi, C/F, L/gallon), and trip checklists.

Education toolkit for DSH agents: lesson-plan skeletons by grade band, quiz scaffolding and answer-key validation, analytic rubrics, Anki/markdown flashcards, and readability levels.

Academic writing toolkit for DSH agents: paper outlines, title and abstract skeletons, GB/T 7714 / APA / MLA citations, phrasing QA, and a pre-submission checklist.

Tracks the agent's state as Yogācāra mental factors over a durable seed store, and feeds it back into the prompt as first-person inclination.

Project memory for coding agents on DeepSeek Harness: similar past edits surface as precedents, out-of-scope edits are blocked, and cognition persists across sessions (47 upstream-verified fixes, gold-free 20-task run).

Organize folders and sessions into projects with shared PROJECT.md memory and AGENTS.md instructions injected into every session, plus a project tree sidebar.

PostgreSQL-backed durable memory for DSH agents: MCP memory tools plus a Web panel for reviewing memories and candidates.

Focus board for DeepSeek Harness agents: durable, model-maintained notes in the session workspace that pin the objective, constraints, and decisions across compaction and sessions, with automatic context injection, archive on clear, and an optional web panel.

Manage DSH rules from the web settings panel — edit global/project AGENTS.md, or auto-split pasted rules into layered AGENTS.md and skills via LLM.

DeepSeek Harness용 데스크톱 스타일 프로젝트 및 AI 워크벤치

DeepSeek Harness용 페일클로즈드(fail-closed) 방식의 턴별 단계·도구 호출·프로바이더 토큰 예산 제한