mini-SWE-agent integration

mini-SWE-agent support is beta.

tokenjuice install mini-swe-agent writes .mini-swe-agent/tokenjuice.yaml in the current git/project root. This is a YAML configuration fragment for mini-SWE-agent's observation template.

mini-SWE-agent recursively merges multiple config files when passed with -c. Because setting -c replaces the default config list, load the default config explicitly before the tokenjuice fragment:

mini -c mini.yaml -c .mini-swe-agent/tokenjuice.yaml

Install

tokenjuice install mini-swe-agent
tokenjuice doctor mini-swe-agent
tokenjuice uninstall mini-swe-agent

Set MINI_SWE_AGENT_PROJECT_DIR=/path/to/workspace to override the target workspace explicitly.

Behavior

• The fragment keeps mini-SWE-agent command execution unchanged.
• Short observations are rendered as normal.
• Long observations keep deterministic head/tail slices and add guidance to retry noisy commands through tokenjuice wrap -- <command>.
• The fragment tells the agent to use tokenjuice wrap --raw -- <command> only when raw bytes are required.

Current beta caveat

This is not a shell hook and does not run tokenjuice automatically. It is a config-fragment bridge for mini-SWE-agent workflows that want the agent to learn from long-output failures and rerun noisy commands through tokenjuice.