[micahjsmith]

Stacked PRs:

• docs(autoland): add the autoland plan for landing the step-abstraction stack #76
• feat(autoland): add an x step that runs a command and stops the land if it fails #75
• refactor(autoland): inject services everywhere and test with fakes instead of patches #74
• refactor(autoland): handle Ctrl+C once at the run boundary #73
• refactor(autoland): merge land runs as a stack through a plan optimization policy #72
• refactor(autoland): execute each step through a StepContext with injected services #71
• refactor(autoland): have steps describe themselves with a StepView #70
• ->refactor(autoland): give each plan step type its own class #69
• test(autoland): pin step execution behavior with golden scenarios #68

---

Summary

Phase 1 of the autoland step-abstraction stack. Plan steps were plain dataclasses joined by PlanStep = Union[LandStep, WorkflowStep, ConfirmStep], with isinstance chains in every module that handled a plan. This adds an AutolandStep base class and moves the per-type behavior for plan syntax, persistence, and replanning progress onto the step classes.

• New stack_pr.autoland.steps package: base.py (the AutolandStep ABC, a PlanParser context holding the stack, pr_is_merged, the next positional index, and the current line number, and the step registry), plus land.py, workflow.py, and confirm.py. The step classes keep their names and fields and are re-exported from model.py. PlanStep is now an alias of AutolandStep.
• Each step type declares kind (state-file tag), keyword (plan syntax), and label. @register_step adds it to a registry that is looked up by kind or keyword. Registering the same kind or keyword twice raises an error.
• Persistence: to_dict() / from_dict() replace model._STEP_TYPES and the type switch in _deserialize_step. The state file is byte-for-byte unchanged (STATE_VERSION stays 1).
• Plan syntax: parse(args, parser) and plan_line(stack) per step. parse_plan and format_plan_for_editor now dispatch through the registry. The comment-column rule is a per-step aligns_comments flag (false for confirm). A bare w still counts as an unrecognized step, via bare_keyword_ok. PR-reference/URL resolution moves into steps/land.py unchanged.
• Progress: assume_completed(), is_checkpoint_done(), replan_key(landed_prs), credit_from(old), landed_pr(stack), and marks_landed_prefix replace plan._mark_skipped, _checkpoint_done, _credit, and the type switch in _checkpoint_keys. landed_prefix_end, is_assumed_completed, and carry_over_progress stay in plan.py and now call these methods.
• engine.py, display.py, status.py, and native_stack.py keep their isinstance dispatch for now; later phases handle them. display.py and status.py only get assert isinstance(step, ConfirmStep) at their "the remaining type" fallthroughs, because mypy can no longer narrow a base class to the last member of the old union.

Behavior change: a state file step with an unknown type now fails to load with Unknown plan step type: '<type>'. Before, it was silently loaded as a ConfirmStep, which was a latent bug. --resume/--replan already report ValueErrors from loading as "Failed to load state file".

Test plan

• Golden autoland scenarios (tests/test_autoland_golden.py) pass with no golden files regenerated, including the checkpoint JSON
• New tests/test_autoland_steps.py: registry lookup by kind and keyword, duplicate kind/keyword registration error (with no partial registration), unknown state-file step type error, and a round trip of every registered step type through the state file and through plan text (format_plan_for_editor -> parse_plan)
• Full suite passes with GH_CONFIG_DIR empty and GH_TOKEN unset

🤖 Generated with Claude Code

Co-Authored-By: Claude Opus 5.5 (1M context) noreply@anthropic.com