dsh-decisions-classifiers: OpenAI Decisions API Classifiers for DeepSeek Harness
The `dsh-decisions-classifiers` plugin for DeepSeek Harness (DSH) integrates the OpenAI Decisions API to provide agents with structured text classification and decision-making capabilities. It is integrated across four layers: Agent tools, Cordis services, callable gates, and suggestion cards, and includes a built-in classifier library covering harmful content detection, prompt injection defense, dangerous command identification, and intent analysis. Users can flexibly configure the API base URL, model, and thresholds, and enable a pre-execution callable gate mechanism to perform security interception before critical operations, thereby improving agent safety and controllability.
Read Moredsh-web-search-openai: Integrate OpenAI Responses API Web Search for DSH
dsh-web-search-openai is a DSH plugin developed by flg1217, designed to integrate a search provider based on the OpenAI Responses API into `ctx.web` without modifying the DSH source code. The plugin achieves server-side retrieval by invoking the `POST /responses` interface and utilizing the native `web_search` tool, unifying structured results or message-level annotations into citable sources. It supports a Web settings card that allows users to configure the endpoint, model, API Key, and retrieval context; notably, the API Key supports input without echoing or reading from the `OPENAI_API_KEY` environment variable. Installation requires ensuring the DSH version is not lower than 0.1.0-rc.6 and proceeding via `dsh pl`...
Read MoreUse Transcribe Skill to convert meeting recordings and podcasts into searchable text
### 正式翻译： This article introduces OpenAI's official Agent Skill `transcribe`: Based on `gpt-4o-mini-transcribe` and `gpt-4o-transcribe-diarize`, it converts audio and video files into text, with support for speaker diarization and known speaker prompts. It explains the installation directories and activation methods in Codex, Cursor and Claude Code, provides reproducible examples including fast transcription with the bundled CLI and `diarized_json` output, and lists precautions such as the 25MB file size limit, automatic chunking for long audio, and the fact that the diarize model does not support prompts.
Read More