MetaBoard

Coding agents produce something most other tools do not: a durable, inspectable record of how work actually happened. Every retrieval, tool call, and retry lands as a structured event because the executor emits it, not because someone remembered to log it. Content creation has no such record. A task board tells you a draft moved from todo to in_review, but it cannot tell you which twenty articles the outline was derived from, how long the second revision took, or what the rejection note from the editor actually said. Those are the facts you need when a piece underperforms and you want to know why.

MetaBoard runs as a DeepSeek Harness (dsh) plugin and brings the agent execution-trajectory mechanism to non-coding work. It reuses the conversation node assembler in dsh, which is genuinely hard and domain-neutral, folding a raw event stream into materialized business objects and rendering them as a trajectory view. Two interfaces share the same work-item log: a CLI for human operators to create, approve, refile, comment on, and archive work items, and a set of dsh tools for agents to record research material, drafts, revisions, review decisions, and their own reports. The blue edges in the trajectory are not inferred; each tool records its source relationships on its own result, so selecting a row reveals the full payload of sources, draft text, revision notes, and elapsed time.

It is aimed at content teams, editorial workflows, and any multi-step knowledge process where intermediate artifacts matter more than the final state. Keep in mind that MetaBoard is still in its Phase 1 verification stage: the end-to-end loop and the concurrent claim-conflict path have not yet been exercised against a live runtime, so it is not ready for day-to-day use today.