my-opencode-deepseek-config workflow
A ready OpenCode v2 setup on one DeepSeek Flash model: agents, commands and 23 skills, with roles told apart by thinking tier.
- 60 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Code and prompts go to the DeepSeek API under your key.
- The setup allows reading files by default while shell commands need confirmation.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/my-opencode-deepseek-configDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Code and prompts go to the DeepSeek API under your key.
- The setup allows reading files by default while shell commands need confirmation.
This is third-party code. Review the repository files before installing.
What it does
The repository holds a working OpenCode v2 configuration: two primary agents (`orchestrator` and `solo`), ten subagents, 18 commands and 23 skills. The whole system runs on a single model, `deepseek/deepseek-flash`, and roles differ only by thinking tier: none for search and routing, `#low`, `#high` and `#max` for planning, deep work and review. Permissions are strict: reading is allowed by default, shell commands need confirmation, `.env` files are denied, and the `oracle`, `reviewer` and `explore` agents are read-only. Skills such as `code-review`, `codemap`, `git-master`, `security-review` and `spec-workflow` load on demand so they do not occupy context all the time.
Who it is for. OpenCode users who want an economical DeepSeek setup with roles already split instead of an empty config with a single agent.
Good fit when
- You need to run OpenCode on DeepSeek without hand-picking agents, permissions and skills.
- Code needs a deep check and there are separate `/review` and `/deep-review` commands.
- Work follows a spec: proposal, delta specs, tasks and verification in the `spec-workflow` skill.
- A large image or PDF must reach a multimodal model without losing quality, which `vision-prep` handles.
Not a fit when
- You need another model: the setup deliberately uses only one.
- You work in OpenCode 1.18 or older: they are no longer supported.
- You want English commands and docs: the README is written in Chinese.
Example request
Run a deep review of my changes with /deep-review and return findings by file and line.Limitations
The setup targets OpenCode v2: the author verified version 2.2.6 on 9 November 2026 and does not support 1.18.x or older. A DeepSeek API key is needed. Background subagents require `OPENCODE_EXPERIMENTAL_BACKGROUND_SUBAGENTS=true`. The author marks as unverified how reasoning depth and token use change between the `low`, `high` and `max` tiers on Flash. The repository is personal, so it fits as a sample to adapt more than a neutral starter kit.
How to disable. Restore the previous `OPENCODE_CONFIG_DIR` value or unset it and OpenCode reads its usual config again. Delete the cloned folder and remove the DeepSeek key from `auth.json` or the environment.
FAQ
Why only one model?
The author argues Flash speed and price suffice for everyday development, and hard tasks use the `#high` or `#max` tier of the same model. It is a deliberate trade of reasoning ceiling for cost.
How do I connect a DeepSeek key?
Type `/connect` in the TUI, choose DeepSeek and paste the key. Another way is the `DEEPSEEK_API_KEY` variable. The key is stored in `~/.local/share/opencode/auth.json`.
Which commands route to agents?
`/deep` for heavy implementation, `/quick` for single file edits, `/ui`, `/vision`, `/plan`, `/oracle` and `/review`. Each sends the task to the matching agent.
What is set for permissions?
Reading is allowed, shell asks by default, `.env` files are denied and outside folders need confirmation. The `oracle`, `reviewer` and `explore` agents only read.
Is a plugin needed for the skills?
Skills load through the native OpenCode `skill` tool. The only plugin used is `superpowers`, pinned to tag `v7.0.2`.
Discussion
No questions or reviews yet.
Related
WorkflowsAn OpenCode plugin: an orchestrator hands work to six specialists, each on its own model, and merges their results into one answer
A kit that gives OpenCode role-based agents, handoffs, independent review and task state that survives interruptions.
Agents, skills and templates for structured AI development in OpenCode, Codex and Claude Code: docs, plans, review, checkpoints
An OpenCode plugin that runs a team of parallel agents with separate sessions, a shared task board and messaging
An MCP server that runs DeepSeek or your own local model as a sub-agent in Claude Code and Codex with its own tool loop.
A terminal AI coding agent built on opencode: any model with your own keys, subagent swarms, goals and output compression.
README in short
The configuration shows how to build OpenCode v2 around one inexpensive model: roles, permissions, commands and skills in one repository. The author explains the reasons for each choice and flags the unverified spots.