brain_kernel library

Public barrel for the brain_kernel system kernel (carbon copy of brain_kernel for vibe-side iteration).

Headless — no UI dependency. Products (builder, industrial, medical, education, B2B, personal) wire kernel pieces together with their own UI and domain workflow. See docs/00_PRD/PRD.md for the manifesto.

Library dependency policy — the kernel itself does NOT depend on package:mcp_server / package:mcp_client at the type level. It describes tool calls and resource reads through the KernelServerHost / KernelClientHost envelope (in lib/src/system/host/). Hosts that want a mcp_server / mcp_client backed surface import the reference impls via package:brain_kernel/mcp_host.dart. Hosts that drive a custom transport (USB, IPC, in-memory bus) implement the abstracts directly without adding mcp_server / mcp_client to their pubspec.

Classes

AbsolutePeriod: Absolute period - a fixed date range.
ActionDef: Action definition.
AdapterManifest: Manifest describing an IO adapter's identity, capabilities, and constraints.
Agent: Agent — self-contained unit with own LLM conversation context, own model, own forked 4-axis instances, and own growth (FR-FBCORE-AGT-001).
AgentBehaviorConfig: Behavioral limits for an agent's runtime loop.
AgentChatController
AgentChatDispatchResult: Result of one AgentChatController.sendUser call. Surfaces the latest AgentReply plus the turns appended during the dispatch (handy for host-side tool-use loops that need the structured tool-call list without re-reading the whole stream).
AgentConfig: Agent Subsystem configuration. Field defaults match SRS §3-2 CFG-010.
AgentCreatedEvent: agent_created — emitted by AgentFacade.createAgent.
AgentDefinition: A single agent definition. Composes 4-axis knowledge bindings with runtime configuration. The bundle stores the definition only — the host runtime is responsible for instantiating it (resolving the referenced profile / skill / fact / philosophy ids and connecting the model + tools).
AgentDeletedEvent: agent_deleted — emitted by AgentFacade.deleteAgent.
AgentFacade: User-visible Agent Subsystem facade. Created once per KnowledgeSystem and exposed as system.agents.
AgentForkAssignedEvent: agent_fork_assigned — emitted by AgentFacade.assign* after deep copy (or sourceRef registration in copy-on-write mode).
AgentForkEvolvedEvent: agent_fork_evolved — emitted by AgentRegistry.recordEvolution when Growth Tracker detects a mutation in an agent-owned 4-axis instance.
AgentForkSource: Fork from another agent's already-evolved owned instance. This is the "transfer" path — agent B receives the result of agent A's growth as its own starting point and continues evolving from there. axis is included in the encoding so cross-axis confusion is impossible.
AgentGrowth: Agent growth counters — incremented when Growth Tracker detects evolution in agent-owned 4-axis instances. Persisted as part of the Agent model (FR-FBCORE-AGT-004).
AgentInvokedEvent: agent_invoked — emitted by AgentFacade.ask/stream at completion (success or failure). Carries success flag + duration + token usage.
AgentLifecycleFactBuilder: Builds the canonical FactRecords for each lifecycle event so the emit sites stay one-liners. All records use entityId == agentId so FactQuery(entityId: agentId) returns the agent's full timeline in chronological order.
AgentLifecycleFactFailedEvent: agent_lifecycle_fact_failed — emitted when the lifecycle fact mirror (FR-FBCORE-AGT-080..087) fails to write a FactRecord. Hosts listen to this to surface silent fact-write errors that would otherwise be invisible (since recordLifecycleAsFacts is best-effort by design).
AgentLifecycleFactType: Standard fact types emitted by the Agent Subsystem.
AgentLlmSessions: Pool of model-pinned LlmPortAdapter instances. Hosts register an adapter per key (provider id, model id, or composite — host's choice) and read the resulting Map<String, LlmPort> view through providers when wiring InfraPorts.
AgentModelConfig: Model configuration for an agent.
AgentRegistry
AgentReply: Result of an agent invocation (ask).
AgentRuntime
AgentsSection: A section containing agent definitions.
AgentSubsystem: Bundle of components produced by AgentSubsystem.create. Hosts pass the registry and runtime into the wrapping KnowledgeSystem.
AgentToken: One token from stream(...) — emitted incrementally.
AgentToolCall: Provider-agnostic tool invocation captured from an LLM response.
AlwaysCondition: Always true condition.
AnalysisAlert: Evaluated alert result indicating whether a condition was triggered.
AnalysisAlertRuleArtifact: Artifact defining an alert rule with condition and severity.
AnalysisArtifact: Base class for analysis artifacts.
AnalysisArtifactProvenance: Provenance metadata for an analysis artifact.
AnalysisAxisMeta: Axis metadata for chart artifacts.
AnalysisChartArtifact: Artifact containing chart data with axes and multiple series.
AnalysisColumnInfo: Column metadata for an analysis source schema.
AnalysisDataSet: Data set returned from a source query.
AnalysisDataSourcePort: Contract for data source adapters.
AnalysisError: Error that occurred during analysis execution.
AnalysisEvidenceLink: Evidence link referencing source data for a summary artifact.
AnalysisFunctionInfo: Information about a registered analysis function.
AnalysisFunctionPort: Contract for pluggable analysis function registration and execution.
AnalysisFunctionResult: Result from executing an analysis function.
AnalysisInputSource: Definition of an analysis input source.
AnalysisJob: Represents a running or completed analysis job.
AnalysisJobLog: Log entry for an analysis job step execution.
AnalysisMetricArtifact: Artifact containing a single numeric metric value.
AnalysisModelArtifact: Artifact containing a trained or fitted model.
AnalysisOutputDef: Definition of an expected analysis output artifact.
AnalysisParameterSchema: Schema for a function parameter.
AnalysisPort: Abstract port for analysis operations.
AnalysisSeriesArtifact: Artifact containing time-series data points.
AnalysisSourceSchema: Schema describing the structure of an analysis input source.
AnalysisSpec: Complete analysis specification describing inputs, transforms, analysis steps, and expected outputs.
AnalysisSpecMetadata: Metadata associated with an analysis specification.
AnalysisStep: An individual analysis computation step.
AnalysisSummaryArtifact: Artifact containing a textual summary with evidence links.
AnalysisTableArtifact: Artifact containing tabular data.
AnalysisTimePoint: A time-indexed data point in a series.
AnalysisTimeRange: Time range for analysis input data.
AnalysisTransform: A data transformation step applied before analysis.
AppliedIntervention: Individual intervention applied during pipeline processing.
AppraisalEnginePort: Engine contract for appraisal metric computation.
AppraisalMetadata: Metadata about the appraisal computation.
AppraisalPort: Port for appraisal operations over profile dimensions.
AppraisalResult: Complete result of appraisal metric computation.
ApprovalDecision: Individual approval decision.
ApprovalEvent: Event emitted when approval status changes.
ApprovalPort: Port for requesting and managing approvals.
ApprovalRecord: Historical approval record.
ApprovalRequest: Approval request details.
ApprovalResult: Result of an approval request.
ArrayExpr: Array literal expression.
Artifact: Generated artifact from skill execution.
AsrOptions: ASR processing options.
AsrPort: Port for ASR operations.
AsrResult: ASR transcription result.
AsrSegment: Single transcription segment.
Asset: Individual asset definition.
AssetBundle: Asset bundle for grouped assets.
AssetCategoryDescriptor: Resolution result for one AssetCategory.
AssetCategoryMap: Static catalogue. Lookup is O(1).
AssetContent: Asset content.
AssetDirectory: Asset directory definition.
AssetExtractor: Common surface for every asset extractor — host wires the concrete implementation (StubAssetExtractor for now, LlmAssetExtractor later) into KbServerBootstrap.
AssetPort: Port for asset retrieval.
AssetProposal: One pending asset operation. The first cut is "add to a list at the canonical pointer for category" — replace / supersede semantics follow in a later round.
AssetSection: Asset section containing static resources.
AssetTouch: One asset touch — derived from a single pointer in a CanonicalChange.changedPointers list.
AssetTouchObserver: Subscribes to a Canonical and emits one AssetTouch per pointer that can be resolved to a known asset category. Hosts wire this to UI to auto-swap mode/category and scroll-into-view.
AssetValidator: Stateless validator. Holds no canonical reference — the canonical (or a probed copy) is passed in as needed.
AudienceConfig: Configuration for audience adaptation.
AudiencePreferences: Audience preferences.
AuditEvent: Canonical audit event.
AuditFilter: Audit query filter.
AuditPort: Port for audit event recording.
BehaviorDefinition: A single behavior definition (one runnable unit).
BehaviorSection: A section containing behavior definitions.
BehaviorStepDef: A single behavior step — the public step contract { id, do, when, then, dependsOn, onFailure }.
BinaryExpr: Binary expression (+, -, *, /, ==, !=, etc.).
BinaryStoragePort: Port for binary asset storage.
BindingSection: Binding section containing data bindings.
BmConfig: Tunable parameters of the BM25 scoring function.
BmHit: One ranked search hit.
BmIndex: In-memory BM25 index.
Bound: Numeric bound constraint for an argument value.
BoundingBox: Bounding box for visual elements.
BridgeToolDef: Tool definition retained in the bridge's in-process registry — the handler is the runtime callable; description + inputSchema are what an external MCP endpoint (when one is wired) reports to tools/list. Self-describing so hosts can enumerate without a separate metadata table.
BuildArtifact: One emitted artifact.
BuildRequest: User intent for one build run.
BuildResult: Outcome of Converter.run.
Bundle: A bundle containing packaged MCP resources.
BundleActivation: Per-bundle activation: activate / catalog / tear-down standard.
BundleActivationRegistry: Process-singleton multi-instance hub.
BundleActivationResult: Result of an activate call — per-category counts plus per-entry error messages collected during the run.
BundleBuilder: Builder for creating bundles programmatically.
BundleChangeEvent: Event emitted when a bundle changes.
BundleDependency: Dependency on another bundle.
BundleDescriptor: Bundle descriptor used by SkillRegistryPort.
BundleFolder: The twelve reserved folder names in a bundle's .mbd/ tree.
BundleImportSummary: Counts returned by KnowledgeSystem.importBundle for diagnostics.
BundleListing
BundleLoader: Loads and parses MCP bundles from various sources.
BundleManifest: Bundle manifest containing metadata and configuration.
BundleReader
BundleRepository: High-level API for bundle operations.
BundleResource: A resource within a bundle.
BundleResources: Read/write surface for one reserved folder under a bundle's .mbd/ root. All paths are forward-slash separated and resolved relative to the folder root — absolute paths and .. traversal are rejected.
BundleSessionBridge
BundleSigner: Implementations sign arbitrary payload bytes with a single key.
BundleSource: Bundle source information.
BundleSourcePort
BundleStoragePort: Abstract interface for bundle storage operations.
BundleValidator: Validates MCP bundles against schema and rules.
CallExpr: Function call expression.
CandidateRecord: Canonical candidate record.
CandidatesPort: Port for candidate fact operations.
Canonical: In-memory canonical with draft mirror. Single-thread (Flutter UI tick) — concurrent access is queued by callers (see DDD-03 §9).
CanonicalChange: One change event broadcast on Canonical.changes.
Canonicalizer: Canonicalizes JSON data for consistent representation.
CanonicalStoragePort: Strategy that knows how to read, write, and delete a canonical bundle directory. All Canonical disk-side I/O routes through this port — no direct dart:io or McpBundleWriter calls leak into the canonical itself.
CapabilityDescriptor: Descriptor for a device capability (action it can perform).
ChannelAttachment: Attachment in a channel event.
ChannelCapabilities: Capabilities of a channel implementation.
ChannelDeliveryResult: Delivery result for a specific channel.
ChannelDescriptor: Descriptor for a channel within a device (e.g., axis, joint, sensor).
ChannelEvent: Event received from a channel.
ChannelIdentity: Identifies a channel platform (e.g., 'slack', 'telegram', 'http', 'websocket').
ChannelPort: Abstract port for bidirectional channel communication.
ChannelResponse: Response to send to a channel.
ChatLog
ChatSection
ChatSession: Manages a chat session with an LLM, including history and context
ChatTurn: One conversation turn in the assistant pane. Lives in core/types.dart so infra adapters can hydrate it from disk without depending on the feat/chat widget tree.
ChunkingConfig: Chunking configuration.
ChunkMeta: Chunk metadata for multi-part payloads.
Claim: A verifiable assertion - the single canonical type for all packages.
ClaimDef: Claim definition.
ClaimQuery: Query descriptor for ClaimsPort.queryClaims.
ClaimsPort: Port for claim operations.
ClaimValidationEntry: Single-claim validation entry.
ClaimValidationReport: Aggregate validation report.
ClaudeProvider: Implementation of LLM interface for Claude API
CliOriginator
CollectionStoragePort: Abstract collection-based storage port.
Command: Command to execute on a device.
CommandResult: Result of a command execution.
CompatibilityConfig: Compatibility configuration for bundle requirements.
CompositeCondition: Composite condition combining multiple conditions.
ComputedValue: Computed value definition.
ConcurrencyDescriptor: Describes concurrency constraints for an adapter.
ConditionalExpr: Conditional/ternary expression (cond ? then : else).
ConditionEvaluation: Evaluation result for a single policy condition.
Confidence: Utility class for confidence calculations.
ConfidenceThresholdConfig: Configuration for confidence threshold behavior in appraisal.
ConfigPort
Conflict: Conflict detected during execution.
ConflictResolution: Conflict resolution configuration for Ethos merging.
ConflictResolver: Resolves value conflicts between two priorities.
ConsumedOpsPorts: Value type holding only the consumed ports (10 required + 7 optional).
ContentHash: Content hash with algorithm information.
ContentResolver: Resolves content references to actual content.
ContextBudget: Budget constraints for a context bundle.
ContextBundle: Context bundle from FactGraph for skill execution.
ContextBundlePort: Port for context bundle construction.
ContextBundleRequest: Request for ContextBundlePort.buildContextBundle.
ContextClaim: Claim from context.
ContextEntity: Entity from context.
ContextEvent: Event from context.
ContextView: View/summary from context.
ConversationKey: Unique key for a conversation within a channel.
ConversationStore: Storage of per-agent conversation history. TTL sweeper + compression are driven by AgentConfig.
ConversationTurn: One conversation turn (user message + assistant reply pair). Stored per agent in ConversationStore under conv/<agentId>/turns.
Converter: Stateless dispatcher.
CoverageConfig: Coverage configuration.
CustomLlmProvider: Base class for implementing custom LLM providers
CustomLlmProviderFactory: Base factory for custom LLM providers
DataBinding: Data binding definition.
DataSource: Data source definition.
DateRange: Helper class for resolved date ranges.
DecisionEnginePort: Engine contract for decision policy evaluation.
DecisionGuidance: Guidance returned when a policy condition matches.
DecisionModifier: Additional modifier for a decision.
DecisionPort: Port for decision evaluation.
DefaultEthosSeeder: Seeds a default ethos into an EthosStorePort when the store is empty.
DefaultRuntimeContext: Default implementation of RuntimeProfileContext.
DependencyResolver: Resolves bundle dependencies.
DeviceDescriptor: Full descriptor for an IO device.
DeviceMatcher: Rule for matching a device during auto-discovery.
DirectionalAttitude: Fundamental posture toward a domain.
DispatchContext
DispatchSession
DomainAction: domainActions[] entry — discriminated by kind. Either a single-click button (carries tool/icon/tooltip/args) or a selectGroup (carries stateKey/items). Hosts switch on kind to render.
DomainActionSelectItem: kind: "selectGroup" item.
DomainEntry: Result of a DomainStorage.list call.
DomainStorage: Abstract surface — adapters back this with disk / db / graph.
EchoChannelPort: Echo channel port that echoes back messages.
EmbeddedEntity: An embedded entity in the FactGraph.
EmbeddedFact: An embedded fact in the FactGraph.
EmbeddedFactGraphData: Container for embedded FactGraph data.
EmbeddedPolicy: An embedded policy in the FactGraph.
EmbeddedPolicyRule: A rule within an embedded policy.
EmbeddedRelation: An embedded relation in the FactGraph.
EmbeddedSummary: An embedded summary in the FactGraph.
EmbeddingConfig: Embedding configuration.
EmbeddingPlan: Bundle of provider + batch size used by EmbeddingRunner and the CLI / Converter wiring. Lives here so callers don't have to reach into the CLI layer.
EmbeddingProvider
EmbeddingRunner
EmbeddingRunResult
EmergencyStopRequest: Request to perform an emergency stop.
EmergencyStopResult: Result of an emergency stop operation.
EmptyLlmPort: Empty LLM port that throws on use.
EmptyTrustStore: Trust store that rejects every signature.
EnginePorts: Container for the three internal engine contracts plus the optional consumed standard ports (FactsPort/PatternsPort/SummariesPort/LlmPort).
EntitiesPort: Port for entity operations in the fact graph.
EntityQuery: Query descriptor for EntitiesPort.queryEntities.
EntityRecord: Canonical entity record.
EntityTypeDefinition: Entity type definition for FactGraph schema.
EnvelopeMeta: Envelope metadata including capture time and source address.
ErrorHandler: Error handler definition.
ErrorRecoveryHandler: Error recovery handler for bundle loading.
Ethos: Root philosophy definition containing all components.
EthosMetadata: Version, author, timestamps, context metadata for an Ethos.
EthosRecord: Canonical ethos record.
EthosScope: Domain-specific applicability scope for an Ethos.
EthosStorePort: Port for ethos storage.
EvaluationOutput: Evaluation output from profile evaluation.
EvaluationResult: Result of expression evaluation.
EventConfig: Event configuration.
EventPort: Event port for pub/sub.
EvidenceFragment: Evidence fragment.
EvidencePort: Port for evidence extraction operations.
EvolutionProposal: Proposed change to an Ethos based on feedback patterns.
EvolutionRecord: A record of an evolution event for audit and history.
ExecutionBudgetConfig: Execution budget configuration (mirrors mcp_skill ExecutionBudget).
ExecutionMetadata: Execution metadata.
ExpectedResult: Expected result definition.
Expr: Base class for all AST nodes.
ExpressionCondition: Expression-based condition.
ExpressionEnginePort: Engine contract for expression formatting and condition evaluation.
ExpressionEvaluator: Evaluates expressions in a given context.
ExpressionFunctions: Registry of built-in expression functions.
ExpressionPort: Port for expression/template formatting operations.
ExpressionStyle: Complete expression style configuration.
ExprVisitor<T>: Visitor pattern for AST traversal.
ExternalFactGraphRef: External FactGraph reference configuration.
ExtractionConfig: Extraction configuration for FactGraph.
ExtractionRule: Extraction rule for FactGraph.
ExtractionValidator: Validator for extracted FactGraph data.
ExtractorRun: Result of one extractor run — proposals plus an optional human- readable note (e.g. "LLM not configured", "no candidates found").
Fact: A single atomic fact — subject / predicate / object triple with optional confidence and source provenance.
FactFacade: Thin facade over the fact-graph standard ports.
FactGraphConfig: FactGraph configuration.
FactGraphRuntime: Phase 2 composition root for mcp_fact_graph.
FactGraphSchema: Schema definitions for FactGraph entities and relations.
FactGraphSection: A section containing embedded FactGraph data.
FactQuery: Query descriptor for FactsPort.queryFacts.
FactRecord: Canonical fact record used by FactsPort.
FactsPort: Port for fact graph fact operations.
FactsSection: A section containing fact triples.
FactTypeDefinition: Fact type definition for FactGraph schema.
FeatureFlags: Feature flags.
FeedbackEvent: Action result feedback for philosophy evolution.
FileContentResolver: Default file-based content resolver.
FileHash: Individual file hash.
FileStorageAdapter: File system storage adapter implementing BundleStoragePort.
Finding: Finding from rubric evaluation.
FlowBrainRuntimeProbe
FlowBrainWiring
FlowDefinition: Flow definition.
FlowDefinitionWorkflow
FlowError: Structured error for FlowPort operations.
FlowOutput: Flow output definition.
FlowParameter: Flow parameter definition.
FlowPort: Abstract port for bidirectional Flow definition ↔ bundle transformation.
FlowResult<T>: Result wrapper for FlowPort operations.
FlowSection: Flow section containing flow definitions.
FlowStep: Flow step definition.
FlowTrigger: Flow trigger definition.
FlowWriteOutput: Output of the write operation (FlowPort.fromDefinition).
ForkEngine: Performs the fork operation for each of the four axes.
ForkSource: Origin of a fork operation. An agent's owned 4-axis instance can be created from either the workspace pool (a Knowledge Subsystem facade entry) or another agent's already-evolved owned fork. Sealed so the callsite (and pattern matching) makes the choice explicit.
FormatConfig: Configuration for response format.
FormattedResponse: Formatted response from expression processing.
FormAutoFixAction: An auto-fix action applied during validation.
FormAxisConfig: Axis configuration for a chart block.
FormBlock: Abstract base class for content blocks within a form document.
FormCanvasBlock: A live canvas scene embed.
FormChartBlock: A chart visualization block.
FormConditionalBlock: A conditional block that renders different content based on a condition.
FormDataBinding: A data binding that connects a form field to a data source.
FormDocument: A form document instance created from a template.
FormDocumentMetadata: Metadata associated with a form document.
FormDocumentVersion: A version entry in the document history.
FormError: Error details for form operations.
FormFieldBlock: An interactive form field block.
FormFontPolicy: Font policy configuration for form layout.
FormHeadingBlock: A heading content block with level.
FormImageBlock: An image content block.
FormLayoutPolicy: Layout policy for form document rendering.
FormMargins: Margin configuration for form layout.
FormPageSize: Page size configuration for form layout.
FormPatchOperation: A JSON Patch operation for updating a form document.
FormPort: Abstract port for form document operations.
FormRendererMetadata: Renderer metadata and capabilities.
FormRendererPort: Pluggable renderer for multi-format output.
FormRenderMetrics: Rendering performance metrics.
FormRenderOutput: Output from document rendering.
FormRepeatableBlock: A repeatable block that generates multiple instances from a template.
FormResult<T>: Generic result wrapper for form operations.
FormSchema: Schema definition for form validation.
FormSchemaField: A field definition within a form schema.
FormSchemaRule: A validation rule within a form schema.
FormSection: A section within a form document.
FormTableBlock: A table content block with columns and rows.
FormTableColumn: Column definition for a table block.
FormTableRow: A row of data in a table block.
FormTemplate: Form template definition.
FormTemplateDependency: Template dependency entry.
FormTemplateManifest: Template manifest with dependencies and compatibility.
FormTemplatePort: Template storage, retrieval, and version management.
FormTemplateVersion: Template version history entry.
FormTextBlock: A plain text content block.
FormValidationIssue: A validation issue found during document validation.
FormValidationResult: Result of a form document validation.
FourAxisSnapshot: Snapshot of the four-axis material relevant to a single agent turn. Host code populates whichever axes it has loaded for the agent — any axis left empty is silently skipped (the corresponding section header is omitted from the composed prompt).
GoldQuestion: One expected-output question. The bundle "passes" the question when every expectedChunkId appears in the top-K BM25 hits for question, optionally above minRank.
GoldQuestionRunner: Stateless runner — owns no bundle reference; each call rebuilds the BM25 index. A later round will let callers pass a shared BmIndex to avoid the rebuild cost when LiveQueryPreview already keeps one warm.
GoldVerdict: Per-question result.
GrammarConfig: Grammar configuration.
GroupingExpr: Grouping expression (parentheses).
GrowthTracker
HedgingConfig: Configuration for hedging language.
HedgingPhrases: Custom hedging phrases.
HistoryEntry: One row read from history.jsonl.
HistoryLog
HostToolRegistry
HttpAuthConfig: HTTP authentication configuration.
HttpStorageAdapter: HTTP storage adapter implementing BundleStoragePort.
HybridWeights: Hybrid search weights.
IdentifierExpr: Identifier reference (variable name).
ImportOriginator
IndexBuildConfig: Index build configuration.
IndexConfig: Index configuration.
IndexExpr: Index access expression (arrindex).
IndexPort: Port for knowledge index management.
InfraPorts: FlowBrain InfraPorts — KnowledgePorts wrapper that also exposes the llmProviders multi-provider pool used by the Agent Subsystem.
IngestPorts: Bundle of all ingest-related ports.
InMemoryBinaryStoragePort: In-memory binary storage for testing.
InMemoryBundleSource: In-memory BundleSourcePort — useful for tests and hosts that preload every bundle at boot time.
InMemoryCollectionStoragePort: In-memory collection storage for testing.
InMemoryEventPort: In-memory event port for testing.
InMemoryKvStoragePort: In-memory KV storage for testing.
InMemoryStoragePort<T>: In-memory typed storage for testing.
InMemoryTrustStore: Simple in-memory store useful for tests and small hosts.
InProcessKernelServerHost
InstalledBundle: Installed bundle metadata returned by install / list / loadInstalled.
InstallLimits: Upper bounds enforced before and during extraction to defend against malformed or hostile .mcpb payloads.
InstallPolicy: Policy flags for a single install call.
IntegratedAxisEntry: One entry in a workspace's integrated axis listing. Combines pool starters (Knowledge Subsystem seed definitions) and every agent's owned fork into a single union view, so any agent (new or existing) can pick any entry — pool starter or another agent's evolved instance — as the source for a new fork. Returned by agents.listIntegrated(workspaceId, axis).
IntegrityChecker: Integrity checker for MCP bundles.
IntegrityConfig: Integrity configuration for bundle verification.
IntegrityError: Integrity error.
IntegrityResult: Bundle integrity verification result.
Interlock: Interlock that must be satisfied before an action proceeds.
InterpolationExpr: String interpolation expression (${...}).
InterventionEngine: The pipeline intervention engine that applies Philosophy at three stages.
InterventionResult: Result of a pipeline intervention.
IoAuditExportConfig: Configuration for exporting audit records to an external system.
IoAuditPort: Port for recording and querying IO device audit trails.
IoAuditQuery: Filter criteria for querying audit records.
IoAuditRecord: Immutable record of an auditable IO operation.
IoDevicePort: Abstract port for IO device communication.
IoError: Error information from an IO operation.
IoPolicyPort: Port for managing and querying I/O policy rules.
IoRegistryPort: Abstract port for device and adapter registry operations.
IoStreamPort: Abstract port for streaming subscriptions to I/O device topics.
IoStreamSubscription: Active subscription with its associated data stream.
JsonComparator: JSON deep equality comparison.
JsonFileDomainStorage: JSON-file backed adapter. Each namespace owns one document at <rootDir>/<sanitisedNamespace>/state.json — a flat { key: value } map. The file is rewritten atomically (via <file>.tmp rename) on each put / delete so a crashed write can't leave the JSON half-formed.
JsonPatchSet: A full patch set (ordered list of PatchOp).
JudgmentCriterion: Conditional decision rule with preferred action.
KbResourceRef: Parsed kb://<facade>/<id> reference. id is the LOCAL id as the caller wrote it — scopeId has NOT been applied yet. The bridge applies scoping at resolution time so cross-bundle reads (where the caller wrote kb://fact/other_bundle.foo) pass through without rewrite.
KernelApp
KernelClientConnection: A configured outbound connection. Hosts surface one of these per remote server the kernel has been pointed at.
KernelClientHost
KernelContent: Sealed base for the kernel content envelope. Hosts add their own concrete subclasses if a transport supports additional content kinds (the reference mcp implementation maps KernelTextContent → mcp.TextContent and KernelImageContent → mcp.ImageContent).
KernelEndpoint
KernelGetPromptResult: Result of a prompts/get dispatch — the assembled message list the MCP client hands the LLM, plus an optional preamble the host surfaces to the user. Library-independent counterpart of mcp.GetPromptResult.
KernelImageContent
KernelPromptArgument: One argument declared on a kernel-side MCP prompt. Mirrors the wire shape (PromptArgument in mcp_server) without coupling kernel callers to the wire library.
KernelPromptDef: Read-friendly snapshot of one registered prompt. Mirrors KernelToolDef / KernelResourceContent so hosts can list / introspect the prompt surface without holding the live handler closure.
KernelPromptMessage: One message in the assembled prompt response. role matches the MCP role tokens ('user' / 'assistant' / 'system'); content reuses the kernel KernelContent envelope so hosts can return text, images, or future content kinds through the same path the tool surface uses.
KernelReadResourceResult
KernelResourceContent: One resource content entry. Matches the MCP ReadResourceResult element shape (uri + optional text or blob + mimeType).
KernelResourceDescriptor: Resource descriptor returned by resources/list on a remote server.
KernelServerHost
KernelTextContent
KernelToolDef: Tool definition retained by a KernelServerHost so external transports can answer tools/list. Mirrors mcp_server's tool meta without depending on the wire library.
KernelToolDescriptor: Tool descriptor returned by tools/list on a remote server.
KernelToolResult: Result of one tool dispatch. Shape mirrors the MCP CallToolResult envelope (content array + optional isError flag) but is library- independent.
KnowledgeBundleEntry
KnowledgeBundleRegistry
KnowledgeConfig: FlowBrain KnowledgeConfig — mcp_knowledge.KnowledgeConfig wrapper + AgentConfig sub-config.
KnowledgeDocument: Knowledge document.
KnowledgeEvent: Base class for all knowledge events.
KnowledgeEventBus: Event bus for knowledge system.
KnowledgePorts: Flat container of standard ports consumed by KnowledgeSystem.
KnowledgeQueryEngine
KnowledgeQueryHit
KnowledgeSection: Knowledge section containing knowledge sources.
KnowledgeSource: Knowledge source definition.
KnowledgeSourceRef: Knowledge source reference.
KnowledgeSystem: FlowBrain KnowledgeSystem — single entry point for the entire core. Delegates the five Knowledge Subsystem facades to a wrapped mcp.KnowledgeSystem and adds the agents (Agent Subsystem) facade.
KnowledgeWriter
KnowledgeWriteSpec
KvEthosStoreAdapter: Default EthosStorePort implementation backed by a KvStoragePort.
KvIndexCorruptionEvent: kv_index_corruption — emitted by AgentRegistry whenever a _kv index entry (per-axis owned-fork index, or an Agent JSON record) fails to deserialize. The registry recovers gracefully (treats the entry as empty / resets on next write), but hosts that wire this event can surface the corruption to operators or to a diagnostics sink — silent recovery is otherwise invisible.
KvStoragePort: Key-value storage port.
KvStoragePortAdapter
LambdaExpr: Lambda/arrow function expression (x => x * 2).
LanguageConfig: Configuration for language and localization.
LazyForkMaterializedEvent: lazy_fork_materialized — emitted by ForkEngine.materialize when a LazyOwnedFork (stored under forkPolicy: ForkPolicy.copyOnWrite) is converted into an eager OwnedFork. Hosts listen to track the transition cost and to confirm explicit materialize calls completed successfully.
LazyMcpBundle: Lazy bundle that loads sections on demand.
LazyOwnedFork: Lazy fork envelope (FR-FBCORE-AGT-046). Holds only the fork metadata — no resolved payload — and defers the source dereference until the agent actually needs the contents (read for transfer / explicit materialize). Used by ForkEngine when AgentConfig.forkPolicy is ForkPolicy.copyOnWrite.
Lexer: Lexer for the MCP Expression Language.
LifecycleBinding: lifecycle[] entry — binds a host-defined slot (e.g. project.save) to a tool dispatch (optionally with fixed args).
LifecycleHooks: Lifecycle hooks.
LiteralExpr: Literal value (number, string, boolean, null).
LlmCapabilities: LLM capabilities configuration.
LlmChunk: Streaming chunk.
LlmClient: Client for interacting with LLM providers
LlmConfiguration: LLM configuration with enhanced retry capabilities
LLmContent: Base class for all MCP content types
LlmMessage: Represents a chat message in a conversation
LlmOriginator
LlmPort: Abstract LLM Port interface.
LlmPortAdapter
LlmProvider: LLM provider interface for mcp_llm internal use.
LlmProviderFactory: Factory interface for creating LLM providers
LlmRequest: Represents a request to an LLM
LlmResponse: LLM response for completions
LlmTool: Tool definition
LlmToolCall: Tool call request
LlmUsage: Token usage information.
LoaderOptions: Options for loading bundles.
LoadResult<T>: Load result containing the loaded bundle and any issues.
LocalizedInfo: Localized name, description, and icon for a specific locale.
LoggingConfig: Logging configuration.
LogicalExpr: Logical expression (and, or, &&, ||).
ManagerRoutedEvent: manager_routed — emitted by AgentFacade.route.
ManagerRouter
ManifestOnlyCanonicalStorage: Default kernel-side storage. Reads manifest.json directly via McpBundle.fromJson so every typed section round-trips. Writes via McpBundleWriter.writeDirectory (no reserved folders).
MatchedCriterion: Matched judgment criterion result.
McpbPackager
McpBundle: A complete MCP Bundle containing all packaged resources.
McpBundleInstaller: Install lifecycle manager for .mcpb packages.
McpBundleLoader: Main entry point for loading MCP bundles.
McpBundleMutator: Transactional mutation entry point. See library doc for the three guard layers.
McpBundlePacker: Produces deterministic .mcpb bytes from a .mbd/ directory.
McpBundleValidator: Validator for section-based MCP bundles.
McpBundleWriter: Writer entry point for .mbd/ directory trees.
McpClientOriginator
McpLlm: Main class for MCPLlm functionality
McpLoaderOptions: Loader configuration options aligned with design document.
McpPort: Port for Model Context Protocol operations.
McpServerSpec
McpToolRef: MCP tool reference.
McpValidationCodes: Error codes for MCP bundle validation.
MemberExpr: Member access expression (obj.field or obj?.field).
MemorySkillRegistry: In-memory implementation of SkillBundleRegistry.
MemoryStorageAdapter: In-memory storage adapter implementing BundleStoragePort.
MetricCondition: Metric existence/confidence condition.
MetricEvent: Metric event for streaming.
MetricPort: Metric computation port.
MetricResult: Result for a single computed metric.
MetricSpec: Specification for on-demand metric computation.
MetricsPort: Port for evaluation metric operations.
MetricValue: Metric value.
ModelSpec: Per-agent LLM provider+model selector. One agent owns exactly one ModelSpec — provider mixing within an agent is not supported in this spec revision (FR-FBCORE-AGT-003).
MultiLayerContext: Multi-layer context for tension detection.
MutationOutcome<R>: Result returned from a McpBundleMutator.mutate closure.
NavigationConfig: Navigation configuration.
NavigationGuard: Navigation guard.
Notification: Notification to send.
NotificationPort: Port for sending notifications.
NotificationRecord: Historical notification record.
NotificationResult: Result of sending a notification.
NullCoalesceExpr: Null coalescing expression (left ?? right).
NullConfig: No-op ConfigPort for hosts that do not surface configuration.
NullObservability: No-op ObservabilityPort for hosts without telemetry.
NullUiResource: No-op UiResourcePort for hosts without a UI resource surface.
ObjectExpr: Object literal expression.
ObservabilityPort
ObservabilityRecord
OcrOptions: OCR processing options.
OcrPort: Port for OCR operations.
OcrRegion: OCR region with position.
OcrResult: OCR recognition result.
OpsFacade: Thin facade over OpsRuntime / standard ops ports.
OpsPorts: Aggregated port container for mcp_knowledge_ops.
OpsRuntime: High-level facade that assembles the execution engine, scheduler, runbook executor, audit recorder, and the five provided-port adapters.
OutputSchema: Output schema definition.
OwnedFork<T>: Envelope written into AgentRegistry per fork. Holds the original domain object + fork provenance.
PageDefinition: Page definition.
ParameterDef: Parameter definition.
ParameterSchema: Parameter schema for inputs/outputs.
Parser: Parser for the MCP Expression Language.
PartialValidators: Facade exposing per-entry partial validation for every knowledge-category section. See library doc for semantics.
PassagePosition: Position of passage in source document.
PatchApplied
PatchOp: One JSON Patch operation.
PatchOriginator: Typed origin descriptor (DDD-04 §2). Sealed so consumers can switch exhaustively in their UI / audit code.
PatchPipeline: Coordinates patch application against a Canonical.
PatchRejected
PatchResult: Outcome of PatchPipeline.apply.
PatternQuery: Query descriptor for PatternsPort.queryPatterns.
PatternRecord: Canonical pattern record.
PatternsPort: Port for pattern operations.
PayloadEnvelope: Envelope wrapping a typed payload with URI and metadata.
PayloadSource: Source information for a typed payload.
Period: Base class for Period types.
Philosophy: A single philosophy — guiding principle with examples.
PhilosophyConfig: Philosophy configuration.
PhilosophyEngine: Sole PhilosophyPort implementation in mcp_philosophy.
PhilosophyEvaluationContext: Context for philosophy evaluation.
PhilosophyEvaluator: The core evaluation engine that applies an Ethos against a runtime context to produce actionable PhilosophyGuidance.
PhilosophyExample: A single example or counter-example illustrating a philosophy.
PhilosophyFacade: Thin facade over PhilosophyPort / PhilosophyEngine.
PhilosophyGuidance: Result of evaluating an Ethos against a context.
PhilosophyPort: Abstract port interface for Philosophy operations.
PhilosophySection: A section containing philosophy definitions.
PipeExpr: Pipe expression (value | filter).
PipelineConfig: Pipeline configuration.
PipelineContext: Pipeline data for intervention.
PipelineEntry: A single pipeline definition.
PipelinePort: Port for pipeline execution.
PipelineRunHandle: Pipeline run handle.
PipelinesSection: A section containing pipeline definitions.
PlatformRequirements: Platform requirements for the bundle.
Policy: A policy definition with rules.
PolicyCondition: Base class for policy conditions.
PolicyConstraints: Constraints applied when a policy rule allows an action.
PolicyDecision: Result of evaluating a policy against an action.
PolicyEvaluationResult: Result of policy evaluation.
PolicyRule: A single policy rule.
PolicySection: Policy section containing bundle policies.
PolicyTrace: Trace of policy evaluation for a command.
PoolForkSource: Fork from a Knowledge Subsystem pool entry (the workspace seed definition). poolId is the same id used by the underlying facade — e.g. 'editor-default' for system.skill/system.profile/system.philosophy, or a query hash for system.facts.
PortEvent: Event data.
Prefs: File-backed preferences. Atomic write (temp + rename) so a crash mid-flush cannot corrupt the live file (NFR-PERSIST-001).
PrefsSnapshot: Immutable snapshot a host can read/replace as a unit.
Procedure: A procedure defines the execution steps of a skill.
ProcedureStep: Procedure step.
Profile: Represents an AI profile/persona definition.
ProfileApplicationMetadata: Metadata for profile application execution per design/03-runtime.md §5.
ProfileApplicationResult: Complete profile application result per design/03-runtime.md §5.
ProfileBuilder: Fluent builder for creating profiles.
ProfileConfig: Profile configuration.
ProfileContentSection: A content section within a profile.
ProfileDefinition: A profile definition within a bundle.
ProfileExecutionMetadata: Metadata for profile application execution.
ProfileFacade: Thin facade over profile.ProfileRuntime.
ProfileManifest: Profile manifest schema.
ProfileOutput: Complete profile application result per mcp_profile design/03-runtime.md §5.
ProfileRegistry: Registry for managing profiles.
ProfileRuntime: Profile application runtime per docs/03_DDD/core-runtime.md v0.2.0.
ProfileSection: Profile section.
ProfilesSection: A section containing profile definitions.
ProfileSummariesPort: Port for profile summary retrieval.
ProfileSummaryResult: Evaluation-level profile summary result.
Prohibition: Absolute boundary with severity classification.
ProhibitionCheck: Individual prohibition check result.
ProhibitionCheckRequest: Request to check prohibitions.
ProhibitionCheckResult: Result of prohibition checks.
ProhibitionCheckResults: Helper for creating ProhibitionCheckResult with computed violation data.
ProhibitionException: Exception condition for a prohibition.
Project: Lifecycle owner for one .kbproj/ directory. Holds the live Canonical, persisted ProjectMeta, and three sidecars (prefs.json / chat.jsonl / history.jsonl).
ProjectMeta: Metadata persisted to <kbproj>/project.json.
PromptFact: Plain text fact tuple supplied by the host. Kept minimal — facts in mcp_fact_graph are richer, but the prompt only needs a one-liner for each fact, so the host extracts what it wants and feeds it here.
PromptTemplate: MCP prompt template.
PropertyDefinition: Property definition for entities and relations.
ProposedChange: Specific change details for an evolution proposal.
ProviderOptions: Configuration options for LLM providers
PublisherInfo: Publisher identity information.
QueryFilter: Query filter for collection storage.
RateLimit: Rate limit constraint for action invocations.
ReadOptions: Options for a read operation.
ReadResult: Result of a read operation.
ReadResultItem: Single item in a read result.
ReadSpec: Specification for a read operation.
ReadTiming: Timing information for a read operation.
RefreshConfig: Refresh configuration.
RegistryEvent: Event emitted by the registry when device or adapter state changes.
ReinforcementEngine: Engine that analyzes feedback to detect reinforcement patterns and propose philosophy evolution.
ReinforcementPattern: A detected reinforcement pattern from feedback history.
RelationTypeDefinition: Relation type definition for FactGraph schema.
RelativePeriod: Relative period - a duration from reference time.
ReporterConfig: Reporter configuration.
RequiresSection
RerankConfig: Reranking configuration.
ResolutionOption: Available resolution option for a tension.
ResourceDescriptor: Descriptor for a resource exposed by a channel.
ResourceInfo: Resource information.
RetrievalPort: Port for knowledge retrieval operations.
RetrievalResult: Retrieval result.
RetrievedPassage: Retrieved passage.
RetrieverConfig: Retriever configuration.
RetrieverInfo: Retriever information.
RetryConfig: Retry configuration.
ReviewerEngine
ReviewerQueue: FIFO queue of AssetProposals. Stateless w.r.t. canonical state — approval is the only path that mutates the bundle, and it goes through the supplied PatchPipeline so dry-run validation, history audit, and undo / redo all stay consistent.
ReviewerVerifiedEvent: reviewer_verified — emitted by AgentFacade.review.
ReviewResult: Reviewer engine result (FR-FBCORE-AGT-025).
RouteDefinition: Route definition.
RoutingDecision: Manager router decision (FR-FBCORE-AGT-024).
RubricCriterion: Rubric criterion.
RubricScore: Rubric score from evaluation.
RunbookDescriptor: Runbook descriptor.
RunbookEntry: A single runbook definition.
RunbookExecution: Result of a runbook execution.
RunbookPort: Port for runbook execution.
RunbooksSection: A section containing runbook definitions.
RunQuery: Query descriptor for RunsPort.queryRuns.
RunRecord: Canonical run record.
RunsPort: Port for run record operations.
RuntimeContextBuilder: Fluent builder for RuntimeProfileContext.
RuntimeDescriptor: Runtime description injected by the host process.
RuntimeProfileContext: Immutable context for profile evaluation.
SchedulerConfig: Scheduler configuration.
ScheduleTarget: Target for a scheduled/triggered action.
ScheduleTriggerPort: Port for scheduled and event-based triggers.
ScoringLevel: Scoring level.
SessionHandle: Lightweight handle for a stream subscription / UI mount / scratch resource owned by a session. The bridge calls close in bulk on session.close() so background streams don't leak when a tab closes or an app deactivates.
SessionRegistry
SettingsAction: settings[] entry — row rendered in the settings dialog footer area. Distinct from the form-schema SettingsSection; this is the trigger-action row.
SettingsField: One form field inside a settings section. Field types are host-defined; common values seen in workspace bundles: text, bool, menu, folder, file, number. Hosts MAY add more.
SettingsSection
SettingsSectionEntry
Signature: Digital signature for bundle.
SignedPayloadRef: Reference to signed payload.
SkillAction: Action from skill execution.
SkillBundle: A skill bundle containing procedures and resources.
SkillBundleRegistry: Abstract registry for runtime use.
SkillConfig: Skill configuration.
SkillFacade: Thin facade over skill.SkillRuntime.
SkillManifest: Skill manifest - Identity and metadata.
SkillModule: Skill module definition.
SkillOutput: Skill output definition.
SkillParameter: Skill parameter definition.
SkillPorts: Container for the ports required by SkillRuntime and friends.
SkillProcedure: Skill procedure definition.
SkillRegistryPort: Port for skill bundle registry.
SkillResult: Standard result from skill execution.
SkillRubric: Skill rubric for evaluation.
SkillRunHandle: Handle for an executing or completed skill run.
SkillRuntime: Main runtime for executing skills.
SkillRuntimePort: Port for skill execution.
SkillSection: Skill section containing skill modules.
SkillStep: A step within a skill.
SkillTrigger: Skill trigger definition.
SlashCommand: Discriminated entry: either template-mode (pre-fill composer) or tool-mode (direct dispatch). Real bundles carry the dispatch either inline (tool + args) or nested (directDispatch.{tool, args}) — the loader accepts both, the emitter writes the canonical inline form.
SourceInfo: Describes where a piece of evidence came from.
SourceReference: Source reference for external sources.
SplashConfig: Splash screen configuration.
StateAdjuster: Adjusts PhilosophyGuidance intensity based on state weighting.
StateDefinition: State definition.
StateWeighting: Dynamic adjustment factors for philosophy execution intensity.
StateWeightingImpact: Audit record of how state modified philosophy output.
StepAction: Step action definition.
StepResult: Result from a single step execution.
StorageMetadata: Storage metadata.
StorageOptions: Storage options.
StoragePort<T>: Generic storage port for typed entities.
StubAnalysisDataSourcePort: Stub data source port for testing.
StubAnalysisFunctionPort: Stub function port for testing.
StubAnalysisPort: Stub implementation of AnalysisPort for testing.
StubAppraisalPort: Stub appraisal port for testing.
StubApprovalPort: Stub implementation for testing.
StubAsrPort: Stub ASR port for testing.
StubAssetExtractor: No-op default. Surfaces a clear "not configured" note so the chat / MCP tool can tell the caller the LLM path isn't available yet — the reviewer queue stays untouched.
StubAssetPort: Stub implementation for testing.
StubAuditPort: Stub implementation for testing.
StubCandidatesPort: Stub implementation for testing.
StubChannelPort: Stub channel port for testing.
StubClaimsPort: Stub implementation for testing.
StubContextBundlePort: Stub implementation for testing.
StubDecisionPort: Stub decision port for testing.
StubEntitiesPort: Stub implementation for testing.
StubEthosStorePort: Stub implementation for testing.
StubEvidencePort: Stub evidence port for testing.
StubExpressionPort: Stub expression port for testing.
StubFactsPort: Stub implementation for testing.
StubFlowPort: Stub implementation of FlowPort for testing.
StubFormPort: Stub form port for testing.
StubFormRendererPort: Stub renderer for testing.
StubFormTemplatePort: Stub template port for testing.
StubIndexPort: Stub implementation for testing.
StubIoAuditPort: In-memory stub for testing audit operations.
StubIoDevicePort: Stub implementation of IoDevicePort for testing.
StubIoPolicyPort: Stub implementation of IoPolicyPort for testing.
StubIoRegistryPort: Stub implementation of IoRegistryPort for testing.
StubIoStreamPort: Stub implementation of IoStreamPort for testing.
StubLlmPort: Stub LLM port for testing.
StubMcpPort: Stub MCP port for testing.
StubMetricPort: Stub metric port for testing.
StubMetricsPort: Stub implementation for testing.
StubNotificationPort: Stub implementation for testing.
StubOcrPort: Stub OCR port for testing.
StubPatternsPort: Stub implementation for testing.
StubPhilosophyPort: Stub philosophy port for testing.
StubPipelinePort: Stub implementation for testing.
StubProfileSummariesPort: Stub implementation for testing.
StubRetrievalPort: Stub implementation for testing.
StubRunbookPort: Stub implementation for testing.
StubRunsPort: Stub implementation for testing.
StubScheduleTriggerPort: Stub implementation for testing.
StubSkillRegistryPort: Stub implementation for testing.
StubSkillRuntimePort: Stub implementation for testing.
StubSummariesPort: Stub implementation for testing.
StubUiPort: Stub implementation of UiPort for testing.
StubVisionPort: Stub Vision port for testing.
StubWorkflowPort: Stub implementation for testing.
SubscriptionHandle: Handle identifying an active subscription.
SubscriptionStatus: Status information for an active subscription.
SummariesPort: Port for summary operations.
SummaryRecord: Canonical summary record.
SyncPolicy: Sync policy for external FactGraph.
SystemPromptComposer
Tension: Detected tension between Philosophy and another layer.
TensionDetector: Detects and resolves conflicts between Philosophy and other layers.
TensionResolution: Resolution of a detected tension.
TensionSource: Source of a tension (which layers are in conflict).
TestAssertion: Test assertion.
TestCase: Test case definition.
TestConfig: Global test configuration.
TestFixture: Test fixture definition.
TestSection: Test section containing test definitions.
TestSessionHandle: In-test helper. Production code creates sessions via the BundleSessionBridge.openSession helper, never directly.
TestStep: Test step definition.
TestSuite: Test suite definition.
ThemeConfig: Theme configuration — bundle-side wrapper around the runtime theme JSON.
ThresholdCondition: Threshold-based condition.
Token: A token produced by the lexer.
TokenUsage: Token-usage breakdown returned by the LLM port.
ToneConfig: Configuration for communication tone.
ToolEntry: A single tool declaration.
ToolInfo: Tool information.
ToolResult: Tool execution result.
ToolsSection: A section containing tool entries.
TopicOptions: Options for a topic subscription.
TopicSpec: Specification for a topic subscription.
TransportDescriptor: Describes a transport type and its default configuration.
TriggeredRule: A rule that was triggered during evaluation.
TrustedPublicKey: A public key the host accepts as a bundle signer.
TrustStore: Host-controlled trust anchor.
TypeCoercer: Type coercion utility for parsing values.
TypeCoercionRules: Type coercion rules for flexible parsing.
TypedPayload: Typed payload with value, unit, timestamp, and quality.
UiError: Structured error for UiPort operations.
UiPort: Abstract port for bidirectional UI definition ↔ bundle transformation.
UiResourceEvent
UiResourcePort
UiResult<T>: Result wrapper for UiPort operations.
UiSection: UI section. Typed fields are deprecated; consumers must read the ui/ reserved folder via BundleResources. The original ui: JSON map is preserved verbatim under raw so toJson round-trips.
UiWriteOutput: Output of the write operation (UiPort.fromDefinition).
UnaryExpr: Unary expression (!, -, not).
UndoFrame: Public, serializable snapshot of one stack frame. Used by sidecar persistence (UndoLog) and by hosts rehydrating from disk.
UndoFrameSnapshot: Persisted shape of one _PatchFrame. Public because the snapshot crosses the sidecar boundary.
UndoLog
UndoRedoStack: Bounded LIFO of patch frames.
UndoSnapshot: Snapshot of the entire stack — undo + redo halves, in their natural stack order (top-of-stack last).
UndoState: Snapshot of the stack — emitted on every push / undo / redo / clear so UI buttons stay in sync.
UserOriginator
ValidationContext: Validation context for tracking state during validation.
ValidationIssue: One row in a ValidationReport.
ValidationReport: Aggregated validator output. isValid == errors.isEmpty.
ValidationResult: Result of bundle validation.
ValidationRule: A custom validation rule.
ValidatorOptions: Options for bundle validation.
ValuePriority: Ordered principle for value conflict resolution.
ValueResolution: Value conflict resolution result.
VisionFace: Vision face detection.
VisionLabel: Vision label/object.
VisionOptions: Vision processing options.
VisionPort: Port for vision analysis operations.
VisionResult: Vision analysis result.
VocabularyConfig: Vocabulary configuration.
WidgetDefinition: Reusable widget definition.
WidgetNode: Widget node in the UI tree.
WiringSection
WorkflowDescriptor: Workflow descriptor.
WorkflowEntry: A single workflow definition.
WorkflowPort: Port for workflow execution.
WorkflowRunHandle: Workflow run handle.
WorkflowsSection: A section containing workflow definitions.

Enums

ActionStatus: Action execution status.
ActionType: Action types.
AgentAxis: 4-axis identifier — Knowledge Subsystem domain assigned to an agent via fork (FR-FBCORE-AGT-005).
AgentRole: Agent role — metadata controlling which AgentFacade methods are callable.
AnalysisAlertSeverity: Severity level for analysis alerts.
AnalysisArtifactType: Type of analysis artifact produced.
AnalysisExecutionMode: Execution mode for an analysis job.
AnalysisJobStatus: Status of an analysis job.
AnalysisSourceType: Source type for analysis input data.
AppCategory: App categories for marketplace listing and categorization.
ApprovalDecisionType: Type of decision made by an approver.
ApprovalEventType: Type of approval event.
ApprovalPolicy: Approval policy determines how approvals are collected.
ApprovalPriority: Priority of approval request.
ApprovalStatus: Status of an approval request.
AssertionType: Assertion types.
AssetCategory: Six asset categories produced and edited by knowledge_builder. Each maps 1:1 to an mcp_bundle native section (DDD-13 §2 / SDD §2.2 of MOD-CORE-005 AssetCategoryMap).
AssetType: Asset types.
AttitudeDomain: Domain of attitude applicability.
AudienceContext: Audience context.
AuthType: Authentication types.
BackpressurePolicy: Backpressure handling policy for subscriptions.
BindingDirection: Binding directions.
BuildTarget: Output kinds. Multiple may be requested in one run.
BundleChangeType: Bundle change event types.
BundleMutationReason: Reason discriminator on BundleMutationException.
BundleType: Bundle types.
CandidateStatus: Candidate lifecycle states.
CanonicalChangeKind: Why a CanonicalChange was emitted.
Cardinality: Cardinality constraints for relations.
ChatRole: Role of a single chat turn in the LLM assistant pane (DDD-15 §1).
ChunkingStrategy: Chunking strategies.
ClaimStatus: Verification state of the claim.
ClaimType: Semantic type of the claim content.
CommandStatus: Status of a command execution.
ComparisonOperator: Comparison operators.
CompositeOperator: Composite logical operators.
ConfidenceLevel: Semantic ranges for confidence interpretation.
ConflictStrategy: Strategy for merging conflicting Ethos instances.
ConsumerPackVariant: Shape of the consumer-pack output.
ContentScope: Scope of hash computation.
DataSourceType: Data source types.
Decision: Policy decision outcome.
DecisionAction: Actions that can be recommended by a decision policy.
Directness: Directness levels.
DocumentFormat: Document formats.
DomainActionKind: Discriminator for domainActions[] entries.
EmbeddingProviderKind: Built-in providers we know how to construct from mcp_llm. Voyage AI does not have a first-class mcp_llm provider; it works through the OpenAI-compatible variant if you point openAiCompatibleEndpoint at https://api.voyageai.com/v1.
Empathy: Empathy levels.
EvolutionType: Type of evolution.
Expertise: Expertise levels.
FactGraphMode: Storage mode for FactGraph data.
FeedbackOutcome: Outcome classification for feedback.
ForkPolicy: Fork policy — controls when deep copy occurs at assignment.
Formality: Formality levels.
FormBlockType: Type of content block within a form document.
FormDataSourceType: Source type for form data bindings.
FormDocumentStatus: Status of a form document in its lifecycle.
GrowthKind: Growth event kind — categorizes evolution recorded by Growth Tracker (FR-FBCORE-AGT-006).
HashAlgorithm: Supported hash algorithms.
HedgingLevel: Hedging levels.
HedgingPosition: Hedging position.
HttpAuthType: Authentication types for HTTP requests.
IndexType: Index types.
InstallConflictPolicy: Behaviour when the target installRoot/<id>/ already exists.
IntegrityErrorType: Types of integrity errors.
InterlockAction: Action to take when an interlock condition is met.
InterlockCondition: Condition operator for interlock rules.
InterventionPoint: Pipeline stage for philosophy intervention.
InterventionType: Types of pipeline intervention.
IoAuditType: Type of auditable IO operation.
IoConnectionState: Connection state of an IO device.
JargonLevel: Jargon levels.
KbFacade: Eight kernel-facade categories. The exact set matches the §1 taxonomy.
KernelTransportKind: Transport family the host can ask a KernelServerHost to bind. Hosts that do not implement network transports (in-process only) raise on any value other than inProcess.
KnowledgeSourceType: Knowledge source types.
Length: Response length types.
LoadStrategy: Asset load strategies.
LogLevel: Log levels.
McpServerTransport
MetricSourceType: Source type for metric computation.
ModifierType: Types of decision modifiers.
NavigationType: Navigation types.
NotificationChannel: Channel for notification delivery.
NotificationPriority: Priority of notification.
NotificationStatus: Status of a notification.
NotificationType: Type of notification.
PatternDirection: Direction of a detected reinforcement pattern.
PayloadKind: Kind of payload carried in an envelope.
PayloadRefType: Type of payload reference.
PayloadType: Data type of a typed payload value.
PeriodDirection: Direction of relative period from reference time.
PeriodUnit: Period unit for relative periods.
PolicyAction: Policy action types.
ProhibitionSeverity: Severity of a prohibition: hard = blocking, soft = warning.
ProposalSource: Where a proposal originated. Audit tag — keeps the user-facing log honest about whether a change came from a chat turn, an MCP client, or a batch extractor run.
ProposalStatus: Status of an evolution proposal.
Quality: Quality indicator for a payload value.
RecoveryStrategy: Error recovery strategy.
ReferenceType: Reference types.
RefreshMode: Refresh modes.
RegistryEventType: Type of registry lifecycle event.
ReporterType: Reporter types.
ResolutionStrategy: Strategy for resolving a tension.
ResourceType: Types of resources that can be in a bundle.
ResultType: Result types.
RetrievalMode: Retrieval modes.
ReviewSeverity: Reviewer severity — optional, used when verdict is fail or revise.
ReviewVerdict: Reviewer verdict (parsed from reviewer LLM response).
RunStatus: Run lifecycle status.
SafetyClass: Safety classification for a device capability.
SentenceComplexity: Sentence complexity.
SignatureAlgorithm: Signature algorithms supported.
SkillTriggerType: Skill trigger types.
SourceType: Source type for bundle loading.
StepActionType: Step action types.
StepType: Step types.
Structure: Response structure types.
TensionLayer: Which layer is involved in a tension.
TensionSeverity: Severity of a tension.
TestStepAction: Step actions.
TestType: Test types.
ThresholdOperator: Threshold comparison operators.
TokenType: Token types for the expression language.
ToneConfidence: Tone confidence levels.
ToolKind: Discriminator for how a ToolEntry is dispatched.
ToolScope
TopicMode: Subscription topic delivery mode.
TransportType: Transport family the server should bind to.
TriggerType: Trigger types.
ValidationLayer: Which validator pass produced an issue.
ValidationSeverity: Severity of a ValidationIssue.
VisualPreference: Visual preference.
VoicePreference: Voice preference.

Extensions

BundleExtensions on Bundle: Utility functions for working with bundles.
DecisionActionExtension on DecisionAction
KbFacadeName on KbFacade
ModifierTypeExtension on ModifierType

Constants

bundleSchemaVersion → const String: Version of the bundle schema.
canonicalizer → const Canonicalizer: Default canonicalizer instance.
currentSchemaVersion → const String: Version of the bundle schema.
defaultSchemaVersion → const String: Default schema version.
integrityChecker → const IntegrityChecker: Default integrity checker instance.
jsonComparator → const JsonComparator: Default JSON comparator instance.
keywords → const Map<String, TokenType>: Keywords map.
mcpCapabilityId → const String: Capability namespace — exposed names are mcp.connect, mcp.call_tool, …
supportedSchemaVersions → const List<String>: Supported schema versions.

Properties

io_stdin → Stdin: Indirection so tests can inject their own Stdin. The default value resolves to the real dart:io stdin at lookup time, not at import time.
no setter

Functions

bindUndoLog(UndoRedoStack stack, UndoLog log) → StreamSubscription<UndoState>: Convenience binding: every change emitted by stack flushes a fresh snapshot to log. Returns the subscription so the host can cancel it at dispose time. Reads the in-memory frames each time (cheap — pointer lists), so the disk file always matches the live stack within one stream tick.
canonicalizeJson(dynamic value) → String: Canonicalize JSON value.
clientTools(KernelClientHost host) → Map<String, InProcessToolHandler>: The mcp_client capability handlers keyed by full name (mcp.connect, …), as InProcessToolHandler — the shape a host registers straight into its in-process dispatcher, exactly as it does for standardTools.
cosineSimilarity(List<double> a, List<double> b) → double: Cosine similarity helper.
createEmbeddingProvider({required EmbeddingProviderKind kind, required String model, required String apiKey, String? endpoint}) → EmbeddingProvider: Build an EmbeddingProvider from a provider kind + credentials.
jsonEquals(dynamic a, dynamic b) → bool: Compare JSON values for equality.
pickTransport({TransportType? userOverride, Stdin? stdin}) → TransportType: Resolution rules:
profileToResource(ProfileManifest profile, {String? path}) → BundleResource: Create a BundleResource from a ProfileManifest.
registerClientTools(HostToolRegistry registry, KernelClientHost host) → List<String>: Register the mcp_client capability onto registry under mcp.* for the external endpoint path (wraps each handler with wrapInProcess). Returns the exposed names. Hosts that only use an in-process dispatcher register clientTools directly instead.
sha256Hash(List<int> bytes) → ContentHash: Compute SHA-256 hash of bytes.
sha256HashJson(dynamic value) → ContentHash: Compute SHA-256 hash of JSON value.
sha256HashString(String content) → ContentHash: Compute SHA-256 hash of string.
skillToResource(SkillManifest skill, {String? path}) → BundleResource: Create a BundleResource from a SkillManifest.
standardTools(KernelApp app) → Map<String, InProcessToolHandler>: Compose the seven facade tool maps into one. The returned map is safe to pass to KernelEndpoint.addStandardTools or to consume directly via an in-process dispatcher.
stdErr(String message) → Map<String, dynamic>: Standard error envelope used by every wrapper.
verifyIntegrity(List<int> bytes, String hashString) → bool: Verify content against hash.
wrapInProcess(InProcessToolHandler handler) → KernelToolHandler: Wrap an InProcessToolHandler into a KernelToolHandler. The raw return value is jsonEncoded into one text content; isError is set when the result is a Map carrying ok: false.

Typedefs

BridgeResourceHandler = Future<Object?> Function(String uri): Optional resource handler. Returns the resource content for the given URI; the bridge wraps the call in the current session's Zone so scopeId is applied where relevant.
BridgeResourceServerAdapter = void Function(String uri, String? name, String? description, String? mimeType, BridgeResourceHandler handler): Resource counterpart to BridgeServerAdapter. Lets the host mirror every registerResource onto an external MCP endpoint's resources/list. Called with (uri, name, description, mimeType, handler).
BridgeServerAdapter = void Function(BridgeToolDef def): Optional server-side adapter. Hosts that also expose an external MCP endpoint (vibe_studio's ServerBootstrap.addTool) thread this so every tool the bridge registers also lands on the external transport's tools/list + dispatch path. Hosts that don't expose an endpoint (AppPlayer's client-default mode, headless probes, tests) leave it null and dispatch in-process via the bridge only.
BridgeToolHandler = Future<KernelToolResult> Function(Map<String, dynamic> args): Tool handler signature. The bridge keeps its own in-process registry of these — no specific MCP server library required to dispatch in-process. Hosts that also expose an external endpoint (stdio / streamable HTTP / SSE) thread a server adapter on top of this registry; everyone else just uses the bridge directly.
DispatcherAttach = void Function(String exposedName, KernelToolHandler handler): Callback the host supplies so the registry can attach / detach a handler to its own in-process dispatcher (AppPlayer's ToolDispatcher, vibe_studio's runtime tool executor, a headless in-memory map, etc.). The registry itself owns no dispatcher state.
DispatcherDetach = void Function(String exposedName)
DomainValue = Object?: One stored value inside a namespace — opaque JSON-shaped payload.
EvaluationContext = PhilosophyEvaluationContext: Backward-compatible alias for PhilosophyEvaluationContext.
ExprFunction = dynamic Function(List args): Type alias for expression functions.
InProcessToolHandler = Future<Object?> Function(Map<String, dynamic> args): In-process tool handler. Returns raw JSON-shaped data — the host wraps it for transport when registering with KernelEndpoint.addStandardTools (or via wrapInProcess).
KernelPromptHandler = Future<KernelGetPromptResult> Function(Map<String, dynamic> args): Prompt handler signature used by the kernel core. Receives the caller-supplied arguments (validated upstream against the prompt's declared KernelPromptArgument list) and returns the assembled message list.
KernelResourceHandler = Future<KernelReadResourceResult> Function(String uri, Map<String, dynamic> params): Resource handler signature used by the kernel core.
KernelServerHostFactory = KernelServerHost Function({bool debugMode, required String name, required String version, Set<ToolScope> visibility}): Factory the host supplies to create one KernelServerHost per endpoint. The reference impl McpServerKernelHost.factory (package:brain_kernel/mcp_host.dart) wraps mcp.Server + stdio / Streamable HTTP / SSE transports. Hosts that never expose a network surface leave KernelApp.boot's serverHostFactory null — the kernel falls back to InProcessKernelServerHost for in-process dispatch.
KernelToolHandler = Future<KernelToolResult> Function(Map<String, dynamic> args): Tool handler signature used by the kernel core.
KnowledgeSystemRef = Object Function(): Forward declaration to keep agent_registry.dart decoupled from the concrete KnowledgeSystem wrapper class. The wrapper supplies () => system at wire time.
LlmComplete = Future<String> Function(String prompt)
RuntimeProbeFn = Future<ValidationReport> Function(McpBundle bundle): Callback that probes the runtime layer (DDD-05 §2.3). Defined here so the validator stays decoupled from flowbrain. A typical implementation runs KnowledgeSystem.importBundle in a __validate__ workspace and converts BundleImportSummary / thrown errors into a ValidationReport.
ScreenDefinition = PageDefinition: Backward compatibility alias.
SkillRunner = Future Function(String skillId, Map<String, dynamic> inputs)
SubFlowRunner = Future<Map<String, dynamic>> Function(String flowId, Map<String, dynamic> input)
SystemPromptResolver = Future<String?> Function(String agentId): Strategy for composing the agent's runtime system prompt. Receives the current agent id (host can look up the Agent definition + 4-axis state) and returns the synthesized prompt that will be carried in KnowledgeSystem.agents.ask. Returning null falls back to whatever agent.systemPrompt the registry already holds.
ToolDispatcher = Future Function(String skillId, Map<String, dynamic> inputs): Function signatures the host injects so each step type knows where to dispatch.
UndoRedoEntry = _UndoRedoEntry: Public alias so external callers can reference the entry type without crossing the underscore boundary.

Exceptions / Errors

AgentNotFoundException: Thrown when an agent id passed to ask/stream/assign*/unassign/ update/delete/route/review/getHistory/clearHistory does not resolve to a registered agent (FR-FBCORE-AGT-070).
AgentRoleMismatchException: Thrown when route() is called on a non-manager agent or review() on a non-reviewer agent (FR-FBCORE-AGT-073).
AssetNotFoundException: Asset not found exception.
BundleAlreadyInstalledException: Raised when a bundle with the same id is already installed and the caller requested InstallConflictPolicy.failIfExists.
BundleBusyException: Raised when the install lock cannot be acquired (another installer is running against the same installRoot).
BundleCompatibilityException: Raised when a bundle's CompatibilityConfig is not satisfied by the runtime supplied at install time.
BundleFormatException: Raised when a .mcpb does not start with the ZIP magic bytes or is otherwise not a parseable container.
BundleIntegrityException: Bundle integrity verification exception.
BundleInvalidValueException: Invalid field value error.
BundleLimitException: Raised when a .mcpb exceeds a InstallLimits cap.
BundleLoadException: Base exception for bundle loading errors.
BundleMissingFieldException: Missing required field error.
BundleMutationException: Failure raised from McpBundleMutator.mutate when one of the three transaction guards trips. Reuses BundleWriteException's URI plumbing so callers handling either exception family see a consistent shape.
BundleNotFoundException: Bundle not found exception.
BundleParseException: JSON parsing error with location information.
BundleReadException: Bundle read exception.
BundleReferenceException: Reference validation error.
BundleResourceNotFoundException: Thrown when BundleResources.read / readBytes is called against a file that does not exist under the bundle's reserved folder.
BundleResourceParseException: Thrown when BundleResources.readJson decodes a file that is not valid JSON.
BundleSchemaVersionException: Schema version incompatibility error.
BundleSignatureException: Raised when a declared signature cannot be verified or when an installation policy requires a signature that is missing.
BundleValidationException: Validation exception with multiple errors.
BundleWriteException: Bundle write exception.
ConversationStoreUnavailableException: Thrown when KvStoragePort is not wired but Agent Subsystem is active — surfaced at first ask/stream entry (FR-FBCORE-AGT-072).
EthosValidationException: Thrown when an Ethos structure is invalid.
EvaluationException: Thrown when evaluation logic encounters an error.
EvolutionException: Thrown when evolution proposal generation or application fails.
ForkConflictException: Thrown when a fork is requested for an (agentId, axis, sourceRef) triple that already has an owned instance, and forkPolicy strict mode is enabled (FR-FBCORE-AGT-045, AGT-071).
InterventionException: Thrown when pipeline intervention fails.
LexerException: Lexer exception for invalid input.
ParserException: Parser exception for syntax errors.
PatchApplyException
PhilosophyException: Base exception for all philosophy-related errors.
ProfileNotFoundException: Exception thrown when a profile is not found.
ProhibitionViolationException: Thrown when a hard prohibition is violated.
ProjectFormatException: Thrown when the on-disk layout cannot be parsed as a .kbproj/.
RetrieverNotFoundException: Exception when retriever is not found.
TensionResolutionException: Thrown when tension resolution fails.

brain_kernel library

Classes

Enums

Extensions

Constants

Properties

Functions

Typedefs

Exceptions / Errors

brain_kernel package

brain_kernel library