Feature: Context analysis agent new coverage
  As a developer
  I want the ContextAnalysisAgent workflow thoroughly tested
  So that file loading, dependency analysis, chunking, scoring, and summarization stay stable

  Scenario: Agent initializes with default FakeListLLM
    When I create a ContextAnalysisAgent without an LLM
    Then the agent should use FakeListLLM
    And the agent should have a compiled graph

  Scenario: Agent initializes with custom LLM
    When I create a ContextAnalysisAgent with a mock LLM
    Then the agent should use the provided LLM

  Scenario: _load_files returns preloaded documents when present
    Given a ContextAnalysisAgent instance
    When I invoke _load_files with preloaded documents
    Then the result should return those preloaded documents

  Scenario: _load_files loads real files from disk
    Given a ContextAnalysisAgent instance
    And a temporary Python file on disk
    When I invoke _load_files with the temp file path
    Then the result should contain loaded documents

  Scenario: _load_files reports missing files as errors
    Given a ContextAnalysisAgent instance
    When I invoke _load_files with a nonexistent file path
    Then the result error should mention file not found

  Scenario: _load_files reports non-files as errors
    Given a ContextAnalysisAgent instance
    When I invoke _load_files with a directory path
    Then the result error should mention not a file

  Scenario: _analyze_dependencies extracts deps from documents
    Given a ContextAnalysisAgent instance
    And a state with one loaded document
    When I invoke _analyze_dependencies
    Then the dependencies dict should have entries

  Scenario: _parse_dependencies extracts items from LLM output
    Given a ContextAnalysisAgent instance
    When I parse dependencies from "Dependencies: ['os', 'sys', 'pathlib']"
    Then the parsed list should contain os sys pathlib

  Scenario: _chunk_documents passes small docs through unchanged
    Given a ContextAnalysisAgent instance
    And a state with a small document
    When I invoke _chunk_documents
    Then the chunks should contain the original document

  Scenario: _chunk_documents splits large docs into overlapping chunks
    Given a ContextAnalysisAgent instance with chunk_size 100
    And a state with a large document of 500 characters
    When I invoke _chunk_documents
    Then the chunks should contain more than one document

  Scenario: _score_relevance scores each unique file
    Given a ContextAnalysisAgent instance
    And a state with chunks from two files
    When I invoke _score_relevance
    Then relevance scores should exist for both files

  Scenario: _parse_relevance_score extracts numeric score
    Given a ContextAnalysisAgent instance
    When I parse relevance score from "Score: 0.85 - very relevant"
    Then the parsed score should be approximately 0.85

  Scenario: _parse_relevance_score returns 0.8 for high keyword
    Given a ContextAnalysisAgent instance
    When I parse relevance score from "This is High relevance"
    Then the parsed score should be approximately 0.8

  Scenario: _parse_relevance_score returns 0.3 for low keyword
    Given a ContextAnalysisAgent instance
    When I parse relevance score from "This has low relevance"
    Then the parsed score should be approximately 0.3

  Scenario: _parse_relevance_score returns 0.5 as default
    Given a ContextAnalysisAgent instance
    When I parse relevance score from "No numeric or keyword info"
    Then the parsed score should be approximately 0.5

  Scenario: _summarize_context generates summary
    Given a ContextAnalysisAgent instance
    And a full analysis state with documents and scores
    When I invoke _summarize_context
    Then the direct summary should be a non-empty string

  Scenario: invoke runs the full workflow synchronously
    Given a ContextAnalysisAgent instance
    And a temporary Python file on disk
    When I invoke the full workflow with the temp file
    Then the result should contain summary and scores

  Scenario: stream yields events from the workflow
    Given a ContextAnalysisAgent instance
    And a temporary Python file on disk
    When I stream the workflow with the temp file
    Then at least one analysis event should be yielded

  Scenario: _with_retry wraps a chain that supports retry
    Given a ContextAnalysisAgent instance
    When I wrap a chain with retry support
    Then the wrapped chain should be returned