dispatchlisted
Install: claude install-skill tobyrosen/dispatch-agent
# Dispatch
Hand one self-contained task to a suitable model, inspect its result, and verify the output before integration. Use the primary agent for decisions that require project context or owner judgment.
## Setup
Use Python 3.11 or newer. Follow [README.md](README.md) for installation and backend prerequisites. Keep this skill with the repository so its documentation links resolve; for a Claude skill installation, link the repository directory at `~/.claude/skills/dispatch`.
Run examples from the repository root using its `.venv/bin/python`. Copy `config.example.toml` to `~/.config/dispatch-agent/config.toml` and define aliases under `[models.<alias>]`, with provider defaults under `[providers.<backend>]`. Alternatively, supply `--backend` and a provider model ID in `--model`. Use `--model-id` to override an alias's provider model. CLI options override config; no source edits are needed.
Supported backends are `codex`, `cursor`, `ollama`, `ollama-cloud`, `openrouter`, and `openai-compatible`. Configure HTTP roots with `--base-url` and credential variable **names** with `--api-key-env`. Keep credential values in the environment, out of specs and config. Ollama Cloud defaults to `OLLAMA_API_KEY`; OpenRouter defaults to `OPENROUTER_API_KEY`. Local Ollama at `http://127.0.0.1:11434` needs no key. The generic HTTP backend needs a base URL and defaults to `OPENAI_API_KEY`.
## Model selection and spec discipline
Match task complexity, tool requirements, and output stakes t