Plugins

Falsification-driven correction loop for agent sessions: risky work commits an evidence-checkable claim first, deterministic gates verify it, falsified claims force mutually exclusive replacement hypotheses with discriminating experiments, and every event lands in an append-only evidence ledger.

Evidence-backed, human-approved project memory for DeepSeek Harness

Provider-neutral CI/CD capability for DeepSeek Harness with GitHub Actions support, bounded failure evidence, source-aware diagnostics, and approval-gated rerun/cancel controls.

Clusters arXiv, first-party lab, official artifact, hardware-vendor, and opt-in X signals into versioned release evidence bundles, then gates claim-level reproduction on commands, artifacts, metrics, and verifiers.

Cordis-native, evidence-driven full-stack engineering workflow: inspect-first planning, explicit verification evidence, and approval-aware gating for sensitive operations.

Plug-in vision for text-only models on DSH, with native interaction for image understanding and generation, GUI automation, through layered evidence memory and cache.

Local-first structured vision for text-only agents: images go to a local OpenAI-compatible VLM and come back as JSON evidence (summary, verbatim OCR, layout regions, entities/relations, colors, explicit uncertainty), with anti-hallucination fallback and an optional paste/upload bridge; zero cloud cost, images never leave the machine.

Brings your existing Claude / ChatGPT history into DSH: instant zero-token import of the memories.json inside a Claude export, local Claude Code transcripts with no export needed, and conversations distilled through the model you already configured in DSH. Every memory carries verbatim evidence checked against the source by code — a paraphrase is dropped and counted, and evidence found only in the assistant turns is always routed to review rather than stored.

Local-first failure evidence capsules for DeepSeek Harness sessions

Guides RTL design and verification in DeepSeek Harness, with project scaffolding, evidence checks, and OpenROAD/ORFS setup.

Right-sidebar novel studio that treats a chapter as a pipeline with checkable artefacts instead of one long prompt: scaffold a project, inspect the exact context pack a chapter gets, generate 2-6 parallel candidates and keep the one you pick, then run chapters in sequence with per-chapter progress on disk so it survives an engine restart. Length band, cross-chapter Jaccard and drift ratio are enforced in code rather than by model self-grading; rejections come back as repair evidence for up to two rewrites. AI-fingerprint metrics are reported against a 3956-chapter human baseline but never block, because the distributions overlap. Ships 7 agent tools. The writing engine is a separate Python package that the plugin installs on first use. Needs Python 3.10+ on PATH (or a path set in the plugin settings).

Captures verified project work as masked evidence and reviewable skill proposals; accepted skills become native project skills with version history, diff, rollback, and opt-in global publishing.

Track source lineage for web excerpts and structured Agent reports, fold repeated text with recorded provenance, and export auditable evidence through DSH tools.

Research-style problem solving in four gated stages: record sources by retrieval tier with local before web, turn each shortfall of existing tools into a gap that needs evidence, score candidates against weighted metrics with a deterministic engine, fuse or improve the weak ones, then iterate on measured deltas. Also organizes a directory into human-facing and machine-facing trees.

Test configured DSH models with connectivity, behavioral fingerprint and a local 18-item weighted intelligence battery scoring 0-100 with a tier and per-question evidence, plus animated or custom SVG probes, quick SVG limits, concurrency controls, and latency and token metrics.

Native DeepSeek Harness plugin for evidence-traceable DAG computation and interactive HTML export.

Deep research for DeepSeek Harness: multi-query search, source fetching, evidence extraction and a cited evidence report, through the host ctx.web service. · DSH 深度研究插件：多查询检索、正文取证、交叉印证，产出可引用的证据报告。

Godot game development workbench for DeepSeek Harness. Windows only, preview release: plans a game (GDD, tasks, acceptance checks), changes project files through controlled read/write tools, diagnoses the project and runs it for evidence. Runtime and import checks launch your configured Godot as a normal process with your permissions (the AppContainer sandbox applies only in the restrictedProduction option); the official Godot 4 standard build can be downloaded on request after confirmation (SHA-256 checked), creating a project registers it in Godot's project list, and the selected project context is sent to your configured model.

Evidence-based research workspaces with archived sources, knowledge graphs, exact citations and exportable reports. Experimental alpha. Needs Node 24.9 or later, uv and Python 3.13, plus a one-time setup step (setup-engine.mjs) that copies the engine and runs a locked uv sync; pnpm must also be allowed to run the build scripts of onnxruntime-node and protobufjs.

Multi-workspace agent federation with persistent ticket trees, parallel worker sessions, structured evidence reports, independent acceptance checks, retry/stop and restart recovery. Workers get shell and file tools in the ticket's directory, company_verify re-runs the ticket's acceptance commands in a shell on the host, and scope lists are task authorization rather than a sandbox. State is stored in ~/.dsh/ai-company; the optional company_repo_* tools run separately installed aoci/codegraph commands on any path they are given.

Local-first memory nebula for DeepSeek Harness with graph views, evidence chains, hygiene checks, and offline snapshots. Reads the memory stores of every workspace and of other memory plugins (read-only, for example meow-memory's SQLite) and writes only under $DSH_HOME/memory-nebula. Registers 9 agent tools and a system-prompt section (usage rules every turn, plus not-yet-read neighbouring memories after a few reads); its /export and /share endpoints return the full memory text without authentication on the local server, and an optional standalone engine listens on 127.0.0.1:7890.

Local long-term memory core for DeepSeek Harness with plain-text notes, a bounded index, five tools, optional host-model consolidation and auditable evidence. It appends every user and assistant message under 32,000 characters to ~/.companion/cold/<date>.jsonl in plain text, with no built-in cleanup, and adds a short index of its notes to every system prompt; the optional model-based consolidation is off by default and, once on, sends those messages to the session's model.

Requirement to a detailed design doc another AI can build from: the model fills a plan, the plugin validates it and emits a SPEC plus task cards, then independent review and implementation evidence check the result. Needs Python 3.11 or newer; the spec-graph engine it calls is bundled in the package.

A pet ledger whose achievements come only from real DeepSeek Harness session events: tool-failure streaks, long completed turns, red-to-green test runs, and commit+push success. Every achievement expands into the session id and the actual tool-call evidence (command, result, duration); demo cards are a separate in-memory list with a visible 演示 watermark and never count or persist. Daily quests are generated from the day’s real active time and command types, then frozen. Host + client halves, zero runtime dependencies, 70 tests, local JSON only.