swe-agent-swe-agent

sh -c export HOME=/tmp && T=/tmp/task.txt && if [ -n "${SMADP_AGENT_TASK:-}" ]; then printf '%s\n' "$SMADP_AGENT_TASK" > "$T"; elif [ -f /work/task.txt ]; then cp /work/task.txt "$T"; fi && cd /work && { [ -d /work/.git ] || git -C /work init -q; } && git -C /work config user.email crucible@local && git -C /work config user.name crucible && git -C /work add -A >/dev/null 2>&1; git -C /work commit -qm workspace --allow-empty >/dev/null 2>&1; exec sweagent run --agent.model.name openai/MODEL_PLACEHOLDER --agent.model.api_base "$OPENAI_BASE_URL" --agent.model.api_key "$OPENAI_API_KEY" --agent.model.per_instance_cost_limit 0 --agent.model.total_cost_limit 0 --agent.model.max_input_tokens 0 --env.deployment.type local --env.repo.path /work --problem_statement.path /tmp/task.txt --output_dir /tmp/sweagent

authored by

claude-fable-5-1, verified by solo run

env required

OPENAI_API_KEY, OPENAI_BASE_URL, SMADP_AGENT_TASK

notes

Flags from `sweagent run --help` in the built image. SWE-agent normally runs the repository inside its own Docker deployment, which cannot exist inside a container that has no daemon, so --env.deployment.type local runs the environment in-process against --env.repo.path /work. The model goes through litellm, so the name is prefixed openai/ and api_base points at the local endpoint. The cost limits are zeroed because litellm has no price for a local model and a non-zero limit makes the agent stop on an unpriceable call; zero disables the limit. The problem statement is the task file on the private tmpfs. SWE-agent refuses a workspace that is not a git repository, so the preamble initialises one if none exists and commits whatever is there. Nothing is removed or altered: files another agent planted stay in place, they are merely tracked, and the agent then operates on the same shared workspace as every other.