Skip to content

harness-opencode

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

harness-opencode

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
config
config
 
 
plugins
plugins
 
 
tests
tests
 
 
README.md
README.md
 
 
install.sh
install.sh
 
 
moon.yml
moon.yml
 
 
package.json
package.json
 
 
tsconfig.json
tsconfig.json
 
 

README.md

harness-opencode

opencode at curated pin opencode-ai 1.18.3.

./install.sh   # npm install -g at the pin, config preset + plugins symlinked

What you get

  • config/opencode.json → ~/.config/opencode/opencode.json (only if absent — live config never clobbered): three local providers via @ai-sdk/openai-compatible — ollama, llamacpp, vllm — matching the same ports as the pi and hermes presets. opencode merges configs, so project-level opencode.json can still override.
  • plugins/local-endpoint-status.ts → symlinked into ~/.config/opencode/plugins/: on session start, probes the three endpoints and toasts which are live.
  • Skills: opencode reads agentskills.io skills — install the shared set via packages/skills/install.sh (it targets opencode's skills dir too).

Model selection

Inside opencode: /models → pick Ollama (local) / llama.cpp server / vLLM. Model ids in the preset are placeholders aligned with packages/local-models/MODELS.md — rename to match what you serve.

Verified

  • tsc --noEmit against the pinned @opencode-ai/plugin types — CI-gated.
  • The provider block shape (npm: @ai-sdk/openai-compatible, options.baseURL) follows opencode's documented custom-provider pattern; validate live with opencode → /models after install.