research-session

To set up a reproducible research workspace, you can bootstrap an interactive session that creates a scoped directory with plan, run, and result subfolders. This scaffolding ensures experiment planning and run orchestration remain organized within a single investigation.

Session management for research workflows is the process of creating a scoped workspace with an active-session marker for downstream skills. You need it when running multi-step investigations that require planning, running experiments, and interpreting results reproducibly.

Yes, git worktree isolation works for multiple agents by using the --worktree flag. This gives each session its own worktree and marker, preventing cross-talk when multiple agents run experiments concurrently.

Downstream skills find the active session directory through a written agent_logs/.current file. This marker enables Step-0 session resolution so tools like plan-experiment and run-experiment reliably locate the correct scoped workspace.

You can run the research workspace autonomously or interactively based on the CAUSALAB_AUTONOMOUS setting. The session enforces guardrails that confine writes to the session directory while adjusting behavior for either mode.

The best way to organize experiment planning and results is bootstrapping a session directory with dedicated plan, run, and result subfolders. This structure keeps investigation artifacts separated and reproducible.