Plugins

The pet that grows with your agent: 5 evolution stages earned from real signals (verified rules, completed sessions, tool calls, compactions), agent-state mirroring, and one-click growth share cards. Zero dependencies, 100% local.

Steam-style plugin storefront: a plugin registry browsable inside DSH with one-click install, uninstall and update, plus a zero-dependency CLI and agent-callable market tools.

Scenario bundles of dsh plugins (插件界的 dotfiles): a recipe tool that lists, searches, applies and composes ready-made plugin environments with ordered install sequences.

Cross-ecosystem aggregation data layer: syncs the GitHub dsh-plugin topic and the npm registry into one open JSON index, queried by model tools, a CLI, and a minimal stdio MCP server.

Zero-interruption audit and fuse blocking for DeepSeek Harness: silently records every tool call with deterministic risk scoring, cumulative-risk bonuses, risk-level breakdowns and retention-based cleanup; blocks irreversible catastrophes (protected-path deletion, disk wipe, force-push to protected branches/refs, credential exfiltration), and renders a redacted /risk-guard operation bill with --since filtering.

Agent governance suite: policy-based tool gating (allow/deny/ask with wildcards and priorities), a structured JSONL audit trail, and per-agent token quotas against the host token meter, with state under $DSH_HOME/gov.

Verification-driven self-evolution loop: failure logs become verified AGENTS.md rules; in-session plugin (evolve_learn / evolve_apply / evolve_touch / evolve_recall), tool-verify gate, rule lifecycle, local recall.

S3-compatible backend for the spill seam: oversized tool output goes to object storage (S3, MinIO, R2) with hashed session prefixes and unguessable keys, instead of the agent's local disk.

Complement to official Creator mode: stage, validate, live-mount via ctx.plugin, score, and solidify or roll back in-memory tool-plugin trials.

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.

One-click launcher for the Web UI: a double-click Start-DSH-Web.cmd on the Windows desktop that boots dsh web and auto-opens the browser (polling until ready), plus a web_launcher model tool (install / open / status).

Exports metadata-only DSH turn, model, tool, approval, retry, and error events to Tuning Engines for governed traces, policy evaluation, cost analysis, and Work Session review, with a disk-backed retry queue.

GitHub workbench MCP server (dsh bundle): 23 MCP tools for repos, issues, PRs, code review, and search, with zero runtime dependencies and PAT/OAuth auth.

Disk-persisted execution planning for DSH: milestone/step state machines, dependency gating, an audit event stream, and a deterministic completion gate via dsh tools, CLI, or skill.

Workflow tools for the local aflare binary: generate, validate and run local-first deterministic YAML workflow DAGs (WAL crash recovery, Saga compensation) through the local aflare binary, with 300+ templates built in.

Skill for packaging DeepSeek Harness into a Tauri 2 desktop app (macOS + Windows): dual-platform install playbook, China mirror bootstrap with a subagent timeout sentinel, admin-free Windows toolchain recipe (no VS Build Tools), and three-path acceptance checklists.

Give text-only models vision: a describe_image tool that delegates images to a vision model from your dsh model list over the harness's own LLM runtime.

Hands repetitive text and vision labor (OCR, image analysis, comparison) to a locally running Unsloth Desktop (Unsloth Studio) server through unsloth_run and unsloth_vision tools; pure HTTP client, never spawns or owns processes.

Hands repetitive text and vision labor (OCR, image analysis, comparison) to a local KoboldCpp (llama.cpp) server through koboldcpp_run and koboldcpp_vision tools, with on-demand server lifecycle management.

Transparent image guard for text-only routes: paste images without the 400 session deadlock, plus a vision_analyze tool for OCR/PDF/docx/pptx/video.

Headless-browser verification tools so the agent opens the page it just built, reads the rendered DOM and computed styles, checks the console, and screenshots the result; ships a frontend-verify skill.

Real-browser automation for dsh via an MCP server (Playwright core): navigate, click, fill, type, screenshot, and act on accessibility snapshots with numbered references - all 22 web_* tools re-exposed to the harness over stdio.

Render Markdown + JSON into self-contained HTML documents, cards, dashboards, and galleries (CLI + artifact_render tool).