dsh-codex-bridge

Inside a DeepSeek Harness workflow, agents often need a second opinion or a parallel coding pass from an external agent. Hand-rolling a codex spawn, JSONL capture, polling loop, and output wiring is exactly the kind of repetitive scaffolding a harness plugin should absorb. dsh-codex-bridge does that, turning Codex into a first-class dsh citizen.

Three capabilities carry the weight. First, tool-level invocation: call_codex starts a Codex session in async or block mode with sandbox policy, timeout, and parallelism caps; codex_status polls progress and codex_abort cancels cleanly. Second, same-thread continuation: codex_steer resumes a settled Codex thread with full history, enabling follow-ups, corrections, and multi-turn iteration. Third, live visualization: the Codex tab in the conversation pane renders an Agent Loop waterfall—messages, tool calls with exit codes, turn separators, and the final answer—pushed over the session projection channel and replayable across page refreshes.

Built for developers running DeepSeek Harness who want OpenAI Codex woven into their agent loop. You skip the process-management and UI-polling boilerplate, and instead trigger Codex with a single agent prompt, watch every step unfold in the same panel, and steer the thread in real time.