Plugins

OpenTelemetry (OTLP/HTTP) + Langfuse trace exporter for DeepSeek Harness — GenAI semconv, sanitized before send, batched with retry, off by default.

Record every step the agent took, and walk through it forwards and backwards to find where it went wrong.

Wire-level LLM inspector: captures the raw HTTP request and response bodies of model provider calls and attributes each capture to the session that caused it through the llm/stream waterfall. The request pane renders the body as a collapsible tree with per-node sizes — one measured request was 98.6% a single 973-entry array — and the response pane reassembles the OpenAI-style SSE chunk stream into reasoning, content, tool calls and usage. Shown in a Conversation View tab and a loopback viewer page.

Code graph and knowledge base over an embedded FalkorDB: AST symbols, caller/callee tracing, refactor blast radius, affected-test discovery, clone detection, cross-language porting contracts and spec-to-code traceability.

dsh-plugin: quality-first topic memory for dsh — a topic keeps the starting question, conclusion, impact and dependencies, not the process; local-first git-traced OKF bundle, budgeted LLM-free injection

Source-traceable Brave + Tavily live web search and URL extraction for DeepSeek Harness. Native DSH plugin; TaroCub optional.

The `debug` tool for DeepSeek Harness — a model-facing Debug Adapter Protocol tool with 28 operations (launch/attach, breakpoints, continue/pause/step, threads/stackTrace/scopes/variables/evaluate, disassembly, memory, modules, output, terminate) over the

Runs one coding prompt across multiple configured models in isolated Git worktrees, streams side-by-side tool traces and results, and commits the candidate the user adopts for later rounds.

Side questions (/btw) and ephemeral temporary chat for DeepSeek Harness (dsh) web — ask without interrupting the main agent, dispose without traces

Local JSONL run traces for DeepSeek Harness sessions

Persistent model tool-call trace recorder: every tool call is durably written to a JSONL file that survives restarts, queryable via a structured call_trace tool and a callTraceHistory service, with size rotation and an optional floating canvas UI add-on.

Plataforma de avaliação de agentes: benchmarks em YAML, orquestração de execuções headless, métricas baseadas em traces e relatórios de execução

Ponte para grafo de conhecimento de código do codebase-memory-mcp: busca semântica de símbolos (BM25), trechos de código-fonte, visão geral de arquitetura com comunidades de Leiden, rastreamento de chamadas/fluxo de dados/entre serviços e grep aumentado por grafo