Skip to content

Latest commit

 

History

History

README.md

Examples

Runnable, copy-paste examples for Marshal.

Prerequisites

  • uv sync --extra mcp --extra dev from the repo root.
  • A fleet.config.yaml (copy fleet.config.example.yaml and edit your clients).
  • At least one backend CLI installed and authenticated (e.g. opencode auth login). Marshal does not install or authenticate the backend CLIs for you; run uv run marshal doctor to check.

Files

  • library_quickstart.py — use this when you want the shortest no-driver path: one task, collect the diff, integrate.
  • pipelined_review.py — use this when each implementer job should start its own reviewer as soon as that job finishes (run_many + per-job then).
  • read_paths.py — use this when a run must read a spec or reference file that lives outside its worktree (and to see secret-shaped paths refused).
  • adversarial_review.py — use this when you want a read-only review panel over a run's diff and will decide yourself (the engine computes no verdict).
  • multi_workspace.py — use this when one process should fan work across several repos (list_workspaces / WorkspaceRegistry.run_many).
  • per_client_env.yaml — use this when two clients share a backend but need different provider homes via env: (e.g. two Codex CODEX_HOME values).
  • benchmark-output.md — a captured benchmark + report run: one goal across four clients with a source-honest cost/latency table, and the reasoning behind Marshal's cost-honesty rules.
  • workflows/ — declarative workflow templates (review.yaml, compare.yaml, build-adapters.yaml) you can validate with marshal workflows and run via run_workflow.
  • teams/ — adversarial review team templates (hard-gate.yaml, plan-review.yaml); copy into <repo>/teams/ and point each role at a permission: read-only client.

Run Python examples from the repo root:

uv run python examples/library_quickstart.py
uv run python examples/pipelined_review.py
uv run python examples/read_paths.py
uv run python examples/adversarial_review.py
uv run python examples/multi_workspace.py