cleveractors-core/features/validate_dict.feature

Feature: validate_dict public API — spec-conformant Actor Configuration validation
  As a router platform integrating cleveractors-core
  I want to call validate_dict(config_dict, platform_limits) to validate actor YAML dicts
  So that only spec-conformant configs are accepted before they are stored in the database

  Background:
    Given the validate_dict function is imported from cleveractors

  # ── Top-level key presence ─────────────────────────────────────────────────

  Scenario: Valid minimal config returns the dict unchanged
    Given a minimal valid config dict with agents and routes
    When I call validate_dict with empty platform_limits
    Then the function returns the config dict unchanged
    And no exception is raised

  Scenario: Config missing "agents" key raises ConfigurationError
    Given a config dict without an "agents" key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "agents"

  Scenario: Config missing "routes" key raises ConfigurationError
    Given a config dict without a "routes" key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "routes"

  # ── Agent type validation ──────────────────────────────────────────────────

  Scenario: Config with a valid llm agent type passes validation
    Given a config dict with an agent of type "llm"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Config with a valid tool agent type passes validation
    Given a config dict with an agent of type "tool"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Config with a valid composite agent type passes validation
    Given a config dict with an agent of type "composite"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Config with a valid chain agent type passes validation
    Given a config dict with an agent of type "chain"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Config with a valid template_instance agent type passes validation
    Given a config dict with an agent of type "template_instance"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Config with an unknown agent type raises ConfigurationError
    Given a config dict with an agent of type "unknown_type"
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions agent type validation

  # ── LLM provider validation ────────────────────────────────────────────────

  Scenario: LLM agent with openai provider passes default allowlist validation
    Given a config dict with an llm agent using provider "openai"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: LLM agent with anthropic provider passes default allowlist validation
    Given a config dict with an llm agent using provider "anthropic"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: LLM agent with google provider passes default allowlist validation
    Given a config dict with an llm agent using provider "google"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: LLM agent with unknown provider and no allowlist raises ConfigurationError
    Given a config dict with an llm agent using provider "unknown_provider"
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions provider validation

  Scenario: LLM agent provider matches custom allowed_providers list
    Given a config dict with an llm agent using provider "groq"
    When I call validate_dict with platform_limits containing allowed_providers "groq,openai"
    Then no exception is raised

  Scenario: LLM agent provider not in custom allowed_providers raises ConfigurationError
    Given a config dict with an llm agent using provider "anthropic"
    When I call validate_dict with platform_limits containing allowed_providers "openai"
    Then a ConfigurationError is raised
    And the error message mentions provider validation

  Scenario: LLM agent without explicit provider uses default openai and passes
    Given a config dict with an llm agent without a provider field
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  # ── Route / edge / node / operator validation ──────────────────────────────

  Scenario: Graph route with source/target edge fields passes validation
    Given a config dict with a graph route using source/target edges
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Graph route edge using "from" field raises ConfigurationError
    Given a config dict with a graph route edge using "from" instead of "source"
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions edge field names

  Scenario: Graph route edge using "to" field raises ConfigurationError
    Given a config dict with a graph route edge using "to" instead of "target"
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions edge field names

  Scenario: Graph route with valid node type "agent" passes validation
    Given a config dict with a graph route containing a node of type "agent"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Graph route with invalid node type raises ConfigurationError
    Given a config dict with a graph route containing a node of type "invalid_node"
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions node type validation

  Scenario: Stream route with valid operator type "map" passes validation
    Given a config dict with a stream route containing a "map" operator
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Stream route with invalid operator type raises ConfigurationError
    Given a config dict with a stream route containing an invalid operator type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions operator type validation

  # ── Structural limit enforcement ───────────────────────────────────────────

  Scenario: Graph within max_graph_depth limit passes validation
    Given a config dict with a graph route of 4 edges
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then no exception is raised

  Scenario: Graph exceeding max_graph_depth raises ConfigurationError
    Given a config dict with a graph route of 4 edges
    When I call validate_dict with platform_limits max_graph_depth of 3
    Then a ConfigurationError is raised
    And the error message mentions graph depth

  Scenario: Actor within max_total_nodes limit passes validation
    Given a config dict with a graph route containing 3 non-end nodes
    When I call validate_dict with platform_limits max_total_nodes of 10
    Then no exception is raised

  Scenario: Actor exceeding max_total_nodes raises ConfigurationError
    Given a config dict with a graph route containing 5 non-end nodes
    When I call validate_dict with platform_limits max_total_nodes of 3
    Then a ConfigurationError is raised
    And the error message mentions total nodes

  Scenario: Nested subgraph within max_subgraph_depth limit passes validation
    Given a config dict with nested subgraphs of depth 2
    When I call validate_dict with platform_limits max_subgraph_depth of 3
    Then no exception is raised

  Scenario: Nested subgraph exceeding max_subgraph_depth raises ConfigurationError
    Given a config dict with nested subgraphs of depth 4
    When I call validate_dict with platform_limits max_subgraph_depth of 3
    Then a ConfigurationError is raised
    And the error message mentions subgraph depth

  # ── Return value contract ──────────────────────────────────────────────────

  Scenario: Valid config is returned unchanged (identity contract)
    Given a fully valid spec-conformant config dict
    When I call validate_dict with empty platform_limits
    Then the returned dict is the same object as the input dict

  # ── Callable without file/env/app side effects ────────────────────────────

  Scenario: Bridge route type passes validation
    Given a config dict with a valid bridge route
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  # ── Case-insensitive validation ────────────────────────────────────────────

  Scenario: LLM agent with provider "OpenAI" passes default allowlist
    Given a config dict with an llm agent using provider "OpenAI" in mixed case
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Graph route with node type "Agent" passes validation
    Given a config dict with a graph route containing a node of type "Agent" in mixed case
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  # ── Boundary-exact structural limit tests ──────────────────────────────────

  Scenario: Graph at exact max_graph_depth limit passes validation
    Given a config dict with a graph route of exactly 5 edges for boundary test
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then no exception is raised

  Scenario: Actor at exact max_total_nodes limit passes validation
    Given a config dict with a graph route containing exactly 5 non-end nodes for boundary test
    When I call validate_dict with platform_limits max_total_nodes of 6
    Then no exception is raised

  Scenario: Nested subgraph at exact max_subgraph_depth limit passes validation
    Given a config dict with nested subgraphs of exactly 3 levels for boundary test
    When I call validate_dict with platform_limits max_subgraph_depth of 3
    Then no exception is raised

  # ── Zero-value boundary tests ──────────────────────────────────────────────

  Scenario: max_graph_depth=0 raises ConfigurationError when graph has depth 1
    Given a config dict with a graph route of 1 edge
    When I call validate_dict with platform_limits max_graph_depth of 0
    Then a ConfigurationError is raised
    And the error message mentions graph depth

  Scenario: max_total_nodes=0 raises ConfigurationError when config has any nodes
    Given a config dict with a graph route containing 1 non-end node
    When I call validate_dict with platform_limits max_total_nodes of 0
    Then a ConfigurationError is raised
    And the error message mentions total nodes

  Scenario: max_subgraph_depth=0 raises ConfigurationError when config has a subgraph
    Given a config dict with nested subgraphs of depth 1
    When I call validate_dict with platform_limits max_subgraph_depth of 0
    Then a ConfigurationError is raised
    And the error message mentions subgraph depth

  # ── Empty collection boundary tests ────────────────────────────────────────

  Scenario: Empty agents dict with valid routes passes validation
    Given a config dict with empty agents and valid routes
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Empty routes dict with valid agents passes validation
    Given a config dict with valid agents and empty routes
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Empty allowed_providers list raises ConfigurationError for LLM agent
    Given a config dict with an llm agent using provider "openai"
    When I call validate_dict with platform_limits containing allowed_providers ""
    Then a ConfigurationError is raised
    And the error message mentions provider validation

  # ── Negative platform limits rejected ──────────────────────────────────────

  Scenario: Negative max_graph_depth raises ConfigurationError
    Given a config dict with a graph route of 1 edge
    When I call validate_dict with platform_limits max_graph_depth of -1
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  Scenario: Negative max_total_nodes raises ConfigurationError
    Given a config dict with a graph route containing 1 non-end node
    When I call validate_dict with platform_limits max_total_nodes of -1
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  Scenario: Negative max_subgraph_depth raises ConfigurationError
    Given a config dict with nested subgraphs of depth 1
    When I call validate_dict with platform_limits max_subgraph_depth of -1
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  # ── Crash-bug regression tests ─────────────────────────────────────────────

  Scenario: Non-string LLM provider raises ConfigurationError instead of crashing
    Given a config dict with an llm agent using a non-string provider value
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions non-string provider

  Scenario: Non-string node type raises ConfigurationError instead of crashing
    Given a config dict with a graph route containing a non-string node type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions non-string node type

  Scenario: Edge missing source and target fields raises ConfigurationError
    Given a config dict with a graph route where an edge lacks source and target
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions source and target fields

  Scenario: Graph route missing nodes key raises ConfigurationError
    Given a config dict with a graph route missing the nodes key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "nodes"

  Scenario: Graph route missing edges key raises ConfigurationError
    Given a config dict with a graph route missing the edges key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "edges"

  Scenario: Stream operator missing type field raises ConfigurationError
    Given a config dict with a stream route containing an operator missing type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions missing type

  Scenario: Bare string as allowed_providers raises ConfigurationError
    Given a config dict with platform_limits allowed_providers as a bare string "openai"
    When I call validate_dict with platform_limits containing the bare string
    Then a ConfigurationError is raised
    And the error message mentions allowed_providers type validation

  Scenario: Non-int max_graph_depth raises ConfigurationError
    Given a config dict with platform_limits containing a non-int "max_graph_depth" value
    When I call validate_dict with platform_limits containing a non-int limit value
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  Scenario: agents section that is not a dict raises ConfigurationError
    Given a config dict where agents is a list instead of a mapping
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: routes section that is not a dict raises ConfigurationError
    Given a config dict where routes is a list instead of a mapping
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Agent definition that is not a dict raises ConfigurationError
    Given a config dict where an agent definition is a string instead of a mapping
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Agent with agent_template key but no type skips type validation
    Given a config dict with an agent that has an agent_template key instead of type
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Agent with no type and no template key raises ConfigurationError
    Given a config dict with an agent that has no type and no template key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Agent with template key but no type skips type validation
    Given a config dict with an agent that has a template key instead of type
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: LLM agent with non-dict config raises ConfigurationError
    Given a config dict with an llm agent that has a non-dict config
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions config must be a mapping

  Scenario: Route definition that is not a dict raises ConfigurationError
    Given a config dict where a route definition is a string instead of a mapping
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Route with unknown type raises ConfigurationError
    Given a config dict with a route of unknown type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Graph route with non-list edges raises ConfigurationError
    Given a config dict with a graph route where edges is not a list
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "edges"

  Scenario: Graph route with non-dict edge entry raises ConfigurationError
    Given a config dict with a graph route where one edge is a string
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions edge must be a mapping

  Scenario: Graph route with non-dict node entry raises ConfigurationError
    Given a config dict with a graph route where one node is a string
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Graph route with non-dict nodes section raises ConfigurationError
    Given a config dict with a graph route where nodes is not a dict
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "nodes"

  Scenario: Stream route with non-list operators raises ConfigurationError
    Given a config dict with a stream route where operators is not a list
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Stream route with non-dict operator entry raises ConfigurationError
    Given a config dict with a stream route where one operator is a string
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions operator must be a mapping

  Scenario: Stream operator with empty type string raises ConfigurationError
    Given a config dict with a stream route where one operator has an empty type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions operator type validation

  Scenario: Route with non-dict value raises ConfigurationError before structural limits
    Given a config dict where a route value is not a dict for structural limits
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then a ConfigurationError is raised

  Scenario: Route with non-dict value raises ConfigurationError before total nodes count
    Given a config dict with a non-dict route value for total nodes counting
    When I call validate_dict with platform_limits max_total_nodes of 100
    Then a ConfigurationError is raised

  Scenario: Graph depth raises ConfigurationError when nodes field is not a dict
    Given a config dict with a graph route that has non-dict nodes for depth calculation
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then a ConfigurationError is raised
    And the error message mentions "nodes"

  Scenario: Graph depth check rejects disconnected entry_point
    Given a config dict with a graph route whose entry_point does not match any node
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then a ConfigurationError is raised
    And the error message mentions entry_point not found

  Scenario: Graph depth limit handles cycles gracefully
    Given a config dict with a graph route containing a cycle
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then no exception is raised

  Scenario: Graph depth calculation raises ConfigurationError for non-dict edge entries
    Given a config dict with a graph route where edges list contains a non-dict entry
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then a ConfigurationError is raised
    And the error message mentions edge must be a mapping

  Scenario: Subgraph depth cycle guard prevents infinite recursion
    Given a config dict with two routes that reference each other as subgraphs
    When I call validate_dict with platform_limits max_subgraph_depth of 10
    Then no exception is raised

  Scenario: Subgraph depth raises ConfigurationError for non-existent subgraph reference
    Given a config dict with a subgraph node referencing a missing route
    When I call validate_dict with platform_limits max_subgraph_depth of 5
    Then a ConfigurationError is raised

  Scenario: Graph route with non-dict node definitions raises ConfigurationError before subgraph depth check
    Given a config dict with a graph route containing non-dict node values
    When I call validate_dict with platform_limits max_subgraph_depth of 5
    Then a ConfigurationError is raised

  Scenario: Graph route with non-dict nodes field raises ConfigurationError before subgraph depth check
    Given a config dict with a graph route that has a non-dict nodes field and subgraph limit
    When I call validate_dict with platform_limits max_subgraph_depth of 5
    Then a ConfigurationError is raised
    And the error message mentions "nodes"

  # ── Strengthened error message assertions ──────────────────────────────────

  Scenario: Non-string agent type raises ConfigurationError with specific message
    Given a config dict with an agent that has a non-string type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions non-string agent type

  Scenario: Non-string route type raises ConfigurationError with specific message
    Given a config dict with a route that has a non-string type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions non-string route type

  Scenario: Subgraph node with non-string reference raises ConfigurationError
    Given a config dict with a subgraph node referencing a missing route via a list
    When I call validate_dict with platform_limits max_subgraph_depth of 5
    Then a ConfigurationError is raised

  # ── Cycle 3 fixes: entry_point, edge type, subgraph validation ────────────

  Scenario: Graph route missing entry_point raises ConfigurationError
    Given a config dict with a graph route that has no entry_point field
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "entry_point"

  Scenario: Graph route with non-string entry_point raises ConfigurationError in route validation
    Given a config dict with a graph route whose entry_point is not a string
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "entry_point"

  Scenario: Graph route edge with non-string source raises ConfigurationError
    Given a config dict with a graph route where an edge source is a list
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions source and target fields

  Scenario: Graph route node with empty string subgraph reference raises ConfigurationError
    Given a config dict with a graph route containing a subgraph node with empty reference
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions subgraph reference

  Scenario: Graph route node with non-string subgraph reference raises ConfigurationError
    Given a config dict with a graph route containing a subgraph node with list reference
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions subgraph reference

  Scenario: Route missing type field entirely raises ConfigurationError
    Given a config dict with a route that has no type field
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "type"

  # ── Cycle 4 fixes: empty-string source/target, empty entry_point ──────────

  Scenario: Graph route edge with empty string source raises ConfigurationError
    Given a config dict with a graph route where an edge has an empty source
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions source and target fields

  Scenario: Graph route edge with empty string target raises ConfigurationError
    Given a config dict with a graph route where an edge has an empty target
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions source and target fields

  Scenario: Graph route with empty entry_point string raises ConfigurationError
    Given a config dict with a graph route that has an empty entry_point string
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "entry_point"

  # ── Cycle 4: subgraph pointing to non-graph route ─────────────────────────

  Scenario: Subgraph node referencing a stream route raises ConfigurationError
    Given a config dict with a subgraph node referencing a stream route
    When I call validate_dict with platform_limits max_subgraph_depth of 5
    Then a ConfigurationError is raised

  Scenario: Subgraph node referencing a bridge route raises ConfigurationError
    Given a config dict with a subgraph node referencing a bridge route
    When I call validate_dict with platform_limits max_subgraph_depth of 5
    Then a ConfigurationError is raised

  # ── Coverage gap: non-string stream operator type ─────────────────────────

  Scenario: Stream operator with non-string type raises ConfigurationError
    Given a config dict with a stream route where one operator has a non-string type
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions non-string operator type

  # ── Coverage gap: non-string element in allowed_providers ──────────────────

  Scenario: Allowed_providers list with a non-string element raises ConfigurationError
    Given a config dict with llm agent and allowed_providers containing a non-string element
    When I call validate_dict with platform_limits containing the mixed-type list
    Then a ConfigurationError is raised
    And the error message mentions allowed_providers type validation

  # ── Coverage gap: invalid types on max_total_nodes / max_subgraph_depth ────

  Scenario: Boolean True as max_total_nodes raises ConfigurationError
    Given a config dict with a graph route containing 3 non-end nodes
    When I call validate_dict with platform_limits containing bool max_total_nodes
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  Scenario: Non-int string as max_total_nodes raises ConfigurationError
    Given a config dict with a graph route containing 3 non-end nodes
    When I call validate_dict with platform_limits containing string max_total_nodes
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  Scenario: Boolean True as max_subgraph_depth raises ConfigurationError
    Given a config dict with nested subgraphs of depth 2
    When I call validate_dict with platform_limits containing bool max_subgraph_depth
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  Scenario: Non-int string as max_subgraph_depth raises ConfigurationError
    Given a config dict with nested subgraphs of depth 2
    When I call validate_dict with platform_limits containing string max_subgraph_depth
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  # ── Crash-bug regression: non-string entry_point ────────────────────────────

  Scenario: Non-string entry_point raises ConfigurationError instead of crashing
    Given a config dict with a graph route whose entry_point is not a string
    When I call validate_dict with platform_limits max_graph_depth of 5
    Then a ConfigurationError is raised

  # ── Crash-bug regression: allowed_providers as dict ─────────────────────────

  Scenario: Dict as allowed_providers raises ConfigurationError
    Given a config dict with platform_limits containing a dict for allowed_providers
    When I call validate_dict with platform_limits containing a dict as allowed_providers
    Then a ConfigurationError is raised
    And the error message mentions allowed_providers type validation

  # ── Crash-bug regression: boolean platform_limits ───────────────────────────

  Scenario: Boolean True as max_graph_depth raises ConfigurationError
    Given a config dict with platform_limits containing a bool for max_graph_depth
    When I call validate_dict with platform_limits containing a bool as max_graph_depth
    Then a ConfigurationError is raised
    And the error message mentions platform_limits type validation

  # ── Cycle 5: None argument guards ─────────────────────────────────────────

  Scenario: validate_dict with None config_dict raises ConfigurationError
    When I call validate_dict with config_dict set to None and empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions config_dict and dict

  Scenario: validate_dict with None platform_limits raises ConfigurationError
    Given a minimal valid config dict with agents and routes
    When I call validate_dict with platform_limits set to None
    Then a ConfigurationError is raised
    And the error message mentions platform_limits and dict

  # ── Cycle 5: missing node type key ────────────────────────────────────────

  Scenario: Graph node missing type field produces clear error message
    Given a config dict with a graph route where a node is missing the type key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions missing type

  # ── Cycle 5: ConfigurationError exported from top-level package ───────────

  Scenario: ConfigurationError is importable from the top-level cleveractors package
    When I import ConfigurationError from the cleveractors package
    Then the import succeeds and ConfigurationError is a class

  Scenario: ConfigurationError is listed in cleveractors.__all__
    When I check the cleveractors package __all__
    Then ConfigurationError is listed in the __all__

  # ── Cycle 5: non-dict config_dict and platform_limits ──────────────────────

  Scenario: validate_dict with list config_dict raises ConfigurationError
    When I call validate_dict with config_dict set to an empty list and empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions config_dict and dict

  Scenario: validate_dict with string platform_limits raises ConfigurationError
    Given a minimal valid config dict with agents and routes
    When I call validate_dict with platform_limits set to a string
    Then a ConfigurationError is raised
    And the error message mentions platform_limits and dict

  # ── Cycle 7: DFS step guard branches ──────────────────────────────────────

  Scenario: DFS step guard in _compute_graph_depth raises ConfigurationError
    Given I monkey-patch _MAX_DFS_STEPS to 3
    And a config dict with a graph route of 10 edges
    When I call validate_dict with platform_limits max_graph_depth of 100
    Then a ConfigurationError is raised
    And the error message mentions too complex to validate depth

  Scenario: Subgraph step guard in _compute_subgraph_depth raises ConfigurationError
    Given I monkey-patch _MAX_SUBGRAPH_STEPS to 3
    And a config dict with nested subgraphs of depth 4
    When I call validate_dict with platform_limits max_subgraph_depth of 10
    Then a ConfigurationError is raised
    And the error message mentions too complex to validate

  # ── Cycle 7: Stronger platform_limits error assertions ────────────────────

  Scenario: Negative limit error message mentions "must be >= 0"
    Given a config dict with a graph route containing 1 non-end node
    When I call validate_dict with platform_limits max_total_nodes of -1
    Then a ConfigurationError is raised
    And the error message mentions "must be >= 0"

  Scenario: Type error limit message mentions "must be an integer"
    Given a config dict with a graph route containing 3 non-end nodes
    When I call validate_dict with platform_limits containing string max_total_nodes
    Then a ConfigurationError is raised
    And the error message mentions "must be an integer"

  # ── Cycle 8: Subgraph reference validation without max_subgraph_depth ─────

  Scenario: Subgraph referencing missing route raises error without max_subgraph_depth
    Given a config dict with a subgraph node referencing a missing route without max_subgraph_depth
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Subgraph referencing stream route raises error without max_subgraph_depth
    Given a config dict with a subgraph node referencing a stream route without max_subgraph_depth
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  Scenario: Subgraph referencing bridge route raises error without max_subgraph_depth
    Given a config dict with a subgraph node referencing a bridge route without max_subgraph_depth
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised

  # ── Cycle 8: Edge endpoint validation against declared nodes ─────────────

  Scenario: Edge referencing a non-existent target node raises ConfigurationError
    Given a config dict with a graph route where an edge references a non-existent node
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions edge endpoint not declared

  Scenario: Edge referencing a non-existent source node raises ConfigurationError
    Given a config dict with a graph route where an edge source references a non-existent node
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions edge endpoint not declared

  # ── Cycle 8: LLM provider whitespace stripping ───────────────────────────

  Scenario: LLM agent provider with surrounding whitespace passes validation
    Given a config dict with an llm agent using provider " openai " with whitespace
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  # ── Review fix: reserved agent names (§4.2) ─────────────────────────────

  Scenario: Agent name beginning with double underscore raises ConfigurationError
    Given a config dict with an agent named "__internal_agent"
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions reserved agent name

  # ── Review fix: reserved route names (§5.1) ─────────────────────────────

  Scenario: Route name beginning with double underscore raises ConfigurationError
    Given a config dict with a route named "__input__"
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions reserved route name

  # ── Review fix: route type explicitly set to None ──────────────────────

  Scenario: Route with type set to None raises ConfigurationError
    Given a config dict with a route whose type is null
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions "type"

  # ── Review fix: allowed_providers as set and frozenset ──────────────────

  Scenario: Allowed_providers as a Python set passes validation
    Given a config dict with an llm agent using provider "anthropic"
    When I call validate_dict with an allowed_providers set containing anthropic and openai
    Then no exception is raised

  Scenario: Allowed_providers as a Python frozenset passes validation
    Given a config dict with an llm agent using provider "anthropic"
    When I call validate_dict with an allowed_providers frozenset containing anthropic and openai
    Then no exception is raised

  # ── Review fix: max_total_nodes aggregation across multiple graph routes ──

  Scenario: max_total_nodes aggregates across multiple graph routes
    Given a config dict with two graph routes of 3 nodes each
    When I call validate_dict with platform_limits max_total_nodes of 5
    Then a ConfigurationError is raised
    And the error message mentions total nodes

  Scenario: max_total_nodes at exact limit across multiple graph routes
    Given a config dict with two graph routes of 3 nodes each
    When I call validate_dict with platform_limits max_total_nodes of 6
    Then no exception is raised

  # ── Review fix: allowed_providers as a tuple ────────────────────────────

  Scenario: Allowed_providers as a Python tuple passes validation
    Given a config dict with an llm agent using provider "anthropic"
    When I call validate_dict with an allowed_providers tuple containing anthropic and openai
    Then no exception is raised

  # ── Review fix: empty operators list on stream route ───────────────────

  Scenario: Stream route with empty operators list passes validation
    Given a config dict with a stream route containing no operators
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  # ── Review fix: non-string agent key guard ─────────────────────────────

  Scenario: Non-string key in agents dict raises ConfigurationError
    Given a config dict with a non-string agent key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions agent key must be a string

  # ── Review fix: non-string route key guard ─────────────────────────────

  Scenario: Non-string key in routes dict raises ConfigurationError
    Given a config dict with a non-string route key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions route key must be a string

  # ── Review fix: non-string graph node key guard ──────────────────────

  Scenario: Non-string key in graph route nodes raises ConfigurationError
    Given a config dict with a graph route that has a non-string node key
    When I call validate_dict with empty platform_limits
    Then a ConfigurationError is raised
    And the error message mentions node key must be a string

  # ── Review Cycle 2: LLM default provider exclusion test ──────────────

  Scenario: LLM agent without provider fails when default openai not in allowed_providers
    Given a config dict with an llm agent without a provider field
    When I call validate_dict with platform_limits containing allowed_providers "anthropic,google"
    Then a ConfigurationError is raised
    And the error message mentions provider validation

  # ── Review Cycle 2: positive valid node type coverage ─────────────────

  Scenario: Graph route with valid node type "start" passes validation
    Given a config dict with a graph route containing a node of type "start"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Graph route with valid node type "tool" passes validation
    Given a config dict with a graph route containing a node of type "tool"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Graph route with valid node type "conditional" passes validation
    Given a config dict with a graph route containing a node of type "conditional"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Graph route with valid node type "message_router" passes validation
    Given a config dict with a graph route containing a node of type "message_router"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Graph route with valid node type "function" passes validation
    Given a config dict with a graph route containing a node of type "function"
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  # ── Review Cycle 2: positive valid operator type coverage ─────────────

  Scenario: Stream route with valid operator type "filter" passes validation
    Given a config dict with a stream route containing a "filter" operator
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Stream route with valid operator type "transform" passes validation
    Given a config dict with a stream route containing a "transform" operator
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  Scenario: Stream route with valid operator type "reduce" passes validation
    Given a config dict with a stream route containing a "reduce" operator
    When I call validate_dict with empty platform_limits
    Then no exception is raised

  # ── Review Cycle 2: duplicate subgraph reference de-duplication ───────

  Scenario: Route with duplicate subgraph references does not cause false too-complex error
    Given a config dict with a graph route containing two subgraph nodes referencing the same route
    When I call validate_dict with platform_limits max_subgraph_depth of 5
    Then no exception is raised