OpenCode Processing Skills workflow
Agents, skills and templates for structured AI development in OpenCode, Codex and Claude Code: docs, plans, review, checkpoints
- 61 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The installer copies skills and adapters into agent home directories, and subagents write `docs/`, `plans/` and checkpoint logs.
Install
- Works in
Manual install
git clone git@github.com:DasDigitaleMomentum/opencode-processing-skills.gitThen `cp config.yaml.example config.yaml` and `./install.sh` (global) or `./install.sh --project`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The installer copies skills and adapters into agent home directories, and subagents write `docs/`, `plans/` and checkpoint logs.
This is third-party code. Review the repository files before installing.
What it does
`OpenCode Processing Skills` describes an AI development process where what matters lives in `docs/` and `plans/` files, not in chat history. Skills drive code documentation (`generate-docs`, `update-docs`), multi-session planning with phases and handover (`create-plan`, `resume-plan`, `generate-handover`), independent plan and implementation review and gated implementation with a blueprint before code. The maintainer role delegates to subagents: `delegate`, `retriever`, `doc-explorer`. A separate part, checkpoints: parents and subagents write short done and next steps, failed attempts and approximate input pressure into JSONL files, and a terminal dashboard shows activity. One installer syncs the skills for OpenCode, Codex, Claude Code, Cursor and Hermes.
Who it is for. Teams that run large agent tasks over several sessions and want plan, docs and state kept in the repository.
Good fit when
- A project must be documented from code with symbol inventories so files are not reread
- A large feature is planned in phases with a handover for tomorrow
- An independent subagent reviews the plan and the implementation
- Subagents write checkpoints and a dashboard shows context pressure
Not a fit when
- You need one quick answer with no structure, since the set targets multi-step tasks
- The agent cannot run subagents or work with `docs/` and `plans/`
- You cannot pre-configure models and aliases in `config.yaml`
Example request
Document this project, then draft a multi-phase plan for adding multi-tenancy and save it in `plans/`.Limitations
The author is not affiliated with the OpenCode team. The full set (maintainer, subagents, model aliases, native plugin) works in OpenCode, Codex and Claude Code get the shared skills and checkpoint adapter, and Hermes and Cursor have narrower support. On OpenCode 1.18.2 and newer worker-to-retriever handoff needs `"subagent_depth": 2`. Before upgrading stop the dashboard and every running session. Models are set in `config.yaml`, with an example in `config.yaml.example`.
How to disable. Delete the copied skills and agents from the harness home directory or from `./.opencode/` for a project install, and the `checkpoint-watch` binary in `~/.local/bin` if it was built.
FAQ
How do I install it?
Clone the repository, optionally set models in `config.yaml` and run `./install.sh` for a global install or `./install.sh --project` for `./.opencode/`.
Which agents are supported?
OpenCode with the full set, Codex and Claude Code with shared skills and a checkpoint adapter, Claude Desktop through an MCP adapter, Cursor and Hermes with limits.
What is a checkpoint?
A short agent record of what is done, what is next and how much input context remains. The parent sees headroom and writes a digest or handover in time.
What does the maintainer do?
It holds the direction of work and hands tasks to subagents: `delegate` for research and review, `retriever` for reading files and commands, `doc-explorer` for docs.
Does it depend on a model provider?
No, OpenAI, Anthropic, DeepSeek, Alibaba and others work. Models and aliases are set in `config.yaml`.
Discussion
No questions or reviews yet.
Related
WorkflowsA kit that gives OpenCode role-based agents, handoffs, independent review and task state that survives interruptions.
An open source coding agent for the terminal and desktop with build and plan modes
An OpenCode plugin that runs a team of parallel agents with separate sessions, a shared task board and messaging
A Codex skill collection: phased planning, parallel subagents, documentation access, a browser, 51 hooks and ready agent roles.
A terminal AI coding agent built on opencode: any model with your own keys, subagent swarms, goals and output compression.
Shared infrastructure for Claude Code, Codex, Antigravity CLI and OpenCode: a task lifecycle, skills and sandboxes.
README in short
The documentation explains the approach, shows a quick start, supported harnesses, checkpoints with the Canary dashboard, subagent roles and links to install and module guides.