git:20260816.21ed22e to git:20260902.4a35d69

---

name: opencode-delegation

description: Delegates tasks to OpenCode (opencode run) via delegation-core. Use when delegation-core selects OpenCode or a provider-agnostic harness is wanted.

alwaysApply: false

category: delegation-implementation

tags:

- opencode

- cli

- delegation

- provider-agnostic

dependencies:

- delegation-core

tools:

- opencode

- delegation_executor.py

usage_patterns:

- opencode-cli-integration

- provider-agnostic-routing

complexity: intermediate

model_hint: standard

estimated_tokens: 550

references:

- delegation-core/shared-shell-execution.md

---

# OpenCode Delegation

## Overview

OpenCode is an open-source, provider-agnostic terminal agent published

as npm `opencode-ai`. Its scripting entry point is `opencode run`,

which takes the prompt as positional arguments.

Being provider-agnostic makes it the most flexible entry in the

registry: the model behind it is whatever the local OpenCode config

selects, including a local one.

## When To Use

- `delegation-core` selected `opencode` for the task

- A provider the registry does not model directly is wanted, and

OpenCode already has it configured

- A local or self-hosted model should serve the request

## When NOT To Use

- No provider credentials are configured. `opencode auth list` shows

what is present

- A specific model must be pinned from this side. Model choice belongs

to OpenCode's own config, not to the delegation call

## Prerequisites

### Installation

```bash

npm install -g opencode-ai@latest

```

Homebrew, pnpm, bun, scoop and chocolatey are also documented

upstream.

### Authentication

Credentials resolve per provider through environment variables such as

`ANTHROPIC_API_KEY` or `OPENAI_API_KEY`, through a `.env` file, or

through `opencode auth login`. Stored credentials are written to

`~/.local/share/opencode/auth.json`, and the config file supports

`{env:VARIABLE_NAME}` interpolation.

Verification runs `opencode auth list`, because there is no single

variable to check when the provider set is open-ended.

## Quick Start

### Using the shared delegation executor

```bash

uv run python scripts/delegation_executor.py opencode "Find the bug" \

--files src/

```

### Through the Makefile

```bash

make -C plugins/conjure delegate-opencode PROMPT='Find the bug'

```

### Direct CLI usage

```bash

opencode run "Explain the use of context in this package"

```

## Smart Delegation

`opencode` declares the `code_execution` strength and ranks last among

the network providers in the candidate order, ahead only of the local

`glimmer` service. It declares no model ids, since model choice

belongs to OpenCode's own configuration.

## OpenCode-Specific Details

| Property | Value |

|----------|-------|

| Binary | `opencode` |

| Package | `opencode-ai` |

| Headless form | `opencode run <prompt>` |

| Prompt delivery | positional |

| Version probe | `opencode --version` |

| Auth probe | `opencode auth list` |

`opencode run --file <path>` attaches a file upstream, but this

service inlines file context into the prompt instead, which keeps one

delivery path across every provider that lacks an `@path` syntax.

`opencode serve` starts a headless HTTP server that `opencode run

--attach <url>` can target, and it honors `OPENCODE_SERVER_PASSWORD`.

That path is not modeled here: the registry spawns a process per

delegation rather than holding a server.

An output-format flag is documented for the `session list`, `export`

and `db` subcommands, but not confirmed for `run`, so none is

declared.

## Exit Criteria

- [ ] `opencode` resolves on PATH and answers `opencode --version`

- [ ] `opencode auth list` shows at least one configured provider

- [ ] The built command is `opencode run <prompt>` with the prompt last

and no prompt flag before it

- [ ] File context was inlined into the prompt rather than passed with

`--file`

- [ ] Output saved to `delegations/opencode/YYYYMMDD_HHMMSS.md`

- [ ] This skill ran because `delegation-core` selected `opencode`