AI Agent Hub
Back to plugins
⚙️

dsh-codex-kanban

Workflow Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install sonozaki7/dsh-codex-kanban

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install sonozaki7/dsh-codex-kanban once in DeepSeek Harness to complete the one-time setup; source is at https://github.com/sonozaki7/dsh-codex-kanban

About this plugin

Chat excels at reasoning but is weak at holding operational truth over time. A plausible-sounding answer and a verified delivery look identical inside a conversation window. dsh-codex-kanban makes the lifecycle explicit with six visible states—backlog, ready, running, review, blocked, done—so the gap between planned, edited, tested, and finished is never ambiguous. Acceptance criteria, blockers, execution reports, and verification evidence stay attached to each card and survive every chat session.

The core is a durable, model-callable board stored in a local JSON file. DeepSeek reads and writes the same data through native tools; the kanban_codex tool delegates bounded coding work to Codex inside the current Harness session and records the outcome back onto the card, always landing in review rather than done. The visual workbench binds to 127.0.0.1, supports drag-and-drop on desktop with an inline state menu as a reliable fallback for touch screens and keyboards, uses atomic writes to prevent half-written files, and introduces no account system or cloud dependency.

It is built for developers who juggle multiple repositories and AI chats and want DeepSeek to coordinate while Codex handles scoped implementation; for anyone who needs exact blockers, test results, commits, and URLs kept together on a single card instead of scattered across message history. Local-first, no hidden completion, and no sensitive project context handed to a cloud task service are its ground rules.

Use Cases

  • Juggling multiple repos and AI chats while keeping every task state visible from backlog to done
  • Delegating bounded coding tasks to Codex through DeepSeek, with results always landing in review for human sign-off
  • Keeping blockers, acceptance criteria, and test evidence together on one card instead of lost in chat history

Best For

  • Developers working across multiple repos and AI conversations who need a single view of delivery progress
  • Teams that want DeepSeek to coordinate and Codex to implement scoped tasks with mandatory human review before done
  • Solo developers and small teams who prefer local-first tooling and refuse to hand project context to cloud task services