Feature: Plan Generation Coverage Boost
  Scenarios that exercise previously uncovered defensive code paths
  in the plan_generation module (lines 90-91, 428-429, 491, 631-635).

  Background:
    Given the plan generation coverage module is imported

  Scenario: BoundedMemorySaver prune handles corrupt checkpoint data gracefully
    Given a BoundedMemorySaver with max_checkpoints set to 1
    And the saver storage contains two checkpoints with corrupt data
    When the saver prunes the oldest checkpoint
    Then the prune should complete without error
    And the corrupt checkpoint should be removed from storage

  Scenario: _generate_plan falls back to None when read_text raises an exception
    Given a PlanGenerationGraph instance for coverage testing
    And a generate state with an explicit path pointing to an unreadable file
    When _generate_plan is called with the unreadable file state
    Then a change should be generated with None as original_content
    And the change operation should be MODIFY

  Scenario: _find_matching_context returns None when no context path matches
    Given a PlanGenerationGraph instance for coverage testing
    And a list of contexts with non-matching paths
    When _find_matching_context is called with a path that matches no context
    Then the matching context result should be None

  Scenario: _analyze_contexts falls back when agent.invoke raises an exception
    Given a PlanGenerationGraph instance for coverage testing
    And the ContextAnalysisAgent invoke method is mocked to raise an error
    When _analyze_contexts is called with non-empty contexts
    Then the result should contain the fallback summary
    And the result should contain an analysis error message
    And dependencies and relevance should be empty dicts