Tutorial: agent workflows

Why agent support matters

The package is designed for modern coding agents and token-aware research loops.

Recommended turn split

• Planning turn: use tools, specs, installation matrix, and repo maps.
• Execution turn: run a benchmark or sweep and record artifacts.
• Editing turn: carry only a small patch-oriented context.

Core commands

python -m tscfbench make-agent-spec -o agent_spec.json
python -m tscfbench build-agent-bundle agent_spec.json -o bundle_dir
python -m tscfbench plan-context bundle_dir/manifest.json --phase editing

Why this is useful

The bundle separates compact summaries from larger artifacts so an agent does not need to read every CSV or report in full.