SWE-agent integration

SWE-agent support is beta.

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

SWE-agent accepts multiple --config files and merges them recursively. Load the default config first, then the tokenjuice fragment:

tokenjuice install swe-agent
sweagent run --config config/default.yaml --config .swe-agent/tokenjuice.yaml

commands

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

behavior

• The fragment keeps SWE-agent command execution unchanged.
• It only overrides agent.templates.next_step_truncated_observation_template.
• When SWE-agent reports a clipped observation, the next prompt tells the agent to retry noisy commands with tokenjuice wrap -- <command>.
• tokenjuice wrap --raw -- <command> remains the escape hatch when exact bytes are required.
• The fragment intentionally does not suggest tokenjuice wrap --full.

SWE-agent is maintenance-only upstream and mini-SWE-agent is the newer recommended tool. This integration exists for teams still running SWE-agent configs who want a low-risk tokenjuice guidance bridge.

path override

Set SWE_AGENT_PROJECT_DIR to install or inspect a config fragment for another workspace:

SWE_AGENT_PROJECT_DIR=/path/to/repo tokenjuice install swe-agent