Single-agent baseline that hands the whole search loop to one Claude Code CLI session, run inside the task's own container image by default (subscription-only billing): Claude edits the solution and runs the task's own evaluator, and the framework scores checkpoints with that same evaluator. A run scored on the host (general.eval_mode: local) gets the session as a host subprocess instead.
name: claude_code
display_name: Claude Code
organization: "default"
type: test_time_search
family: single_agent_baseline
summary: "Single-agent baseline that hands the whole search loop to one Claude Code CLI session, run inside the task's own container image by default (subscription-only billing): Claude edits the solution and runs the task's own evaluator, and the framework scores checkpoints with that same evaluator. A run scored on the host (general.eval_mode: local) gets the session as a host subprocess instead."
source: "Galapagos synthesis of SkyDiscover's claude_code controller and Harbor's installed-agent model: the agent and sealed evaluator share the task container, with the agent running non-root and /eval remaining root-owned"
tags: [baseline, single-agent, coding-agent, claude-code, cli, docker, sandboxed]
license: Apache-2.0
controller: galapagos.scaffolds.claude_code.scaffold.ClaudeCodeScaffold
components:
population: {kind: keep_all}
selection_policy: {kind: agent_delegated}
prompt_builder: {kind: one_shot_task_brief}
proposer: {kind: claude_code_cli}
evaluator: {kind: task}
memory: {kind: none}
requirements:
cli: "@anthropic-ai/claude-code on PATH with subscription auth: CLAUDE_CODE_OAUTH_TOKEN (`claude setup-token`) or a `claude /login` session (API-key billing is not supported)"