# Spectre v0.3.0 - Table of Contents

> OTP-native conversational and operational runtime for Elixir agents.

## Pages

- Start here
  - [Spectre](readme.md)
  - [Spectre 0.3.0 guide for LLMs and coding agents](llms.md)
  - [The Spectre System](system.md)
  - [Getting Started](getting_started.md)
  - [Two Realistic Agents](examples.md)
  - [Installation](installation.md)

- Core concepts
  - [Architecture](architecture.md)
  - [Turn semantics and integration boundaries](integrations.md)
  - [DSL](dsl.md)
  - [Canonical Definitions and Audit Projections](canonical_definitions.md)
  - [Definition Store, Resolver, and Manifest V2](definition_store.md)
  - [Stable Identity, Activation, and Definition-Pinned Runs](identity_activation.md)
  - [Event Ownership and Definition Lifecycle](event_lifecycle.md)
  - [Generational Skill State](skill_state.md)
  - [Foundation Conformance](foundation_conformance.md)
  - [Stack](stack.md)
  - [Resumable Runs](runs.md)
  - [Agent Instances and Subjects](instances.md)
  - [Work, Vigil, and the operational runtime](operations.md)
  - [Migrating from 0.1.x to 0.2.0](migrating_to_0_2.md)
  - [Migrating to 0.2.3](migrating_to_0_2_3.md)
  - [Migrating to 0.2.4](migrating_to_0_2_4.md)
  - [Migrating to 0.2.5](migrating_to_0_2_5.md)
  - [Migrating to 0.2.6](migrating_to_0_2_6.md)
  - [Migrating to 0.2.7](migrating_to_0_2_7.md)
  - [Migrating to 0.2.8](migrating_to_0_2_8.md)
  - [Migrating to 0.2.9](migrating_to_0_2_9.md)
  - [Migrating to Spectre 0.3.0](migrating_to_0_3.md)
  - [Skills](skills.md)
  - [Runtime Skills and Routing Projections](runtime_skills.md)
  - [Data-driven execution](data_driven_execution.md)
  - [Governed Definition Changes](governance.md)
  - [Reflective Runtime](reflective_runtime.md)
  - [Public API](api.md)
  - [Spectre public API — 0.3.0](public_api.md)

- Runtime guides
  - [Routing](routing.md)
  - [Routing Evaluation](evaluation.md)
  - [Provider Resilience](providers.md)
  - [Training](training.md)
  - [Actions](actions.md)
  - [Memory](memory.md)
  - [Journal](journal.md)

- Operations
  - [Production Operations](production.md)
  - [Testing](testing.md)
  - [Security Policy](security.md)

- Project
  - [Changelog](changelog.md)
  - [Roadmap](roadmap.md)
  - [Contributing](contributing.md)
  - [LICENSE](license.md)

## Modules

- Canonical definitions and projections
  - [Spectre.Authority.Envelope](Spectre.Authority.Envelope.md): Effective, portable authority granted to one canonical Definition.
  - [Spectre.Canonical.Value](Spectre.Canonical.Value.md): Portable, deterministic value codec used by content-addressed Spectre data.
  - [Spectre.Definition.Candidate](Spectre.Definition.Candidate.md): Immutable Candidate admitted by the activation boundary.
  - [Spectre.Definition.Candidate.Ref](Spectre.Definition.Candidate.Ref.md): Content-addressed identity of a bootstrap Definition Candidate.
  - [Spectre.Definition.Canonical](Spectre.Definition.Canonical.md): Portable, immutable envelope for declared Spectre behavior.
  - [Spectre.Definition.Component](Spectre.Definition.Component.md): One typed component in a canonical Definition envelope.
  - [Spectre.Definition.ContractRegistry](Spectre.Definition.ContractRegistry.md): Trusted registry of component contracts understood by this runtime.
  - [Spectre.Definition.Manifest](Spectre.Definition.Manifest.md): Sealed Manifest V2 for one canonical Definition.
  - [Spectre.Definition.Ref](Spectre.Definition.Ref.md): Content-addressed identity of one canonical Spectre Definition.
  - [Spectre.Definition.Resolver](Spectre.Definition.Resolver.md): Verified resolution of published canonical Definitions.
  - [Spectre.Definition.Store](Spectre.Definition.Store.md): Immutable persistence boundary for canonical Definitions and Manifest V2.
  - [Spectre.Definition.Store.Conformance](Spectre.Definition.Store.Conformance.md): Adapter-neutral Definition Store conformance checks.
  - [Spectre.Definition.Store.Memory](Spectre.Definition.Store.Memory.md): Volatile in-memory reference implementation of `Spectre.Definition.Store`.
  - [Spectre.Execution.Closure](Spectre.Execution.Closure.md): Portable execution dependencies sealed into a Definition Manifest.
  - [Spectre.Foundation.Conformance](Spectre.Foundation.Conformance.md): Executable compatibility gate for Spectre's durable foundations.
  - [Spectre.Projection](Spectre.Projection.md): Deterministic projection interface for canonical Definitions.
  - [Spectre.Projection.Audit](Spectre.Projection.Audit.md): Exact structural projection of a canonical Definition.
  - [Spectre.Projection.Reflection](Spectre.Projection.Reflection.md): Mechanical self-model projection with separate Declared, Effective and
Observed planes.

- Core API
  - [Spectre](Spectre.md): OTP-native conversational runtime for Elixir agents.
  - [Spectre.Agent](Spectre.Agent.md): Declarative DSL for building Spectre agents.
  - [Spectre.AgentRef](Spectre.AgentRef.md): Stable logical identity of a Spectre Agent.
  - [Spectre.Event.Envelope](Spectre.Event.Envelope.md): Portable event envelope whose owner is fixed at Instance admission time.
  - [Spectre.ExternalIdentity](Spectre.ExternalIdentity.md): Authenticated external principal presented by a channel boundary.
  - [Spectre.Instance](Spectre.Instance.md): Subject-scoped OTP owner and fair scheduler for Spectre Runs.
  - [Spectre.Instance.Activation](Spectre.Instance.Activation.md): Canonical snapshot binding one Instance to an active Definition.
  - [Spectre.Instance.Owner](Spectre.Instance.Owner.md): Host contract for canonical Instance ownership and monotonic fencing.
  - [Spectre.Instance.Owner.Lease](Spectre.Instance.Owner.Lease.md): Portable lease issued by an Instance ownership adapter.
  - [Spectre.Instance.Owner.Local](Spectre.Instance.Owner.Local.md): Single-owner lease adapter paired with `Spectre.Instance.Registry`.
  - [Spectre.Instance.Ref](Spectre.Instance.Ref.md): Logical key for one Agent Instance.
  - [Spectre.Instance.Registry](Spectre.Instance.Registry.md): Race-safe local lookup for Agent Instances.
  - [Spectre.Invocation](Spectre.Invocation.md): A concrete, revision-fenced request to execute one staged effect.
  - [Spectre.LinkIntent](Spectre.LinkIntent.md): One bounded, explicit attempt to link a destination identity to a Subject.
  - [Spectre.Reply.Sanitizer](Spectre.Reply.Sanitizer.md): Strips Spectre control tokens from model output before it becomes a
user-visible reply.
  - [Spectre.Run](Spectre.Run.md): Serializable continuation for one unit of Spectre work.
  - [Spectre.Run.Boundary](Spectre.Run.Boundary.md): Observable boundary emitted while advancing a `Spectre.Run`.
  - [Spectre.Run.Ref](Spectre.Run.Ref.md): Opaque, revision-fenced reference to an observable point in a `Spectre.Run`.
  - [Spectre.Run.Request](Spectre.Run.Request.md): Transport-safe description of a Run policy request.
  - [Spectre.Runtime](Spectre.Runtime.md): Turn-level orchestration for Spectre agents.
  - [Spectre.Session](Spectre.Session.md): Conversation-scoped GenServer for running a Spectre agent under supervision.
  - [Spectre.Skill](Spectre.Skill.md): Organizational DSL for reusable, scoped Spectre behavior.
  - [Spectre.Stack](Spectre.Stack.md): Compile-time installer and resolver for Spectre packages.
  - [Spectre.Stack.Conformance](Spectre.Stack.Conformance.md): Executable whole-Stack conformance gate for package ecosystems.
  - [Spectre.Subject](Spectre.Subject.md): Canonical person or entity identity owned by an Agent Instance.
  - [Spectre.Subject.Registry](Spectre.Subject.Registry.md): Authoritative in-memory resolver for Subjects and authenticated identities.
  - [Spectre.SubjectLink](Spectre.SubjectLink.md): Explicit, auditable association between a Subject and an external identity.
  - [Spectre.Supervisor](Spectre.Supervisor.md): Dynamic supervisor for Spectre Sessions and Agent Instances.
  - [Spectre.Turn](Spectre.Turn.md): High-level host-facing turn result.
  - [Spectre.Turn.Dispatcher](Spectre.Turn.Dispatcher.md): Drives a `%Spectre.Turn{}` decision to a delivered outcome.
  - [Spectre.Turn.Handler](Spectre.Turn.Handler.md): Optional boundary through which another runtime may own a normal turn before
Spectre routes it.
  - [Spectre.Turn.Handler.Reply](Spectre.Turn.Handler.Reply.md): Safe, typed reply returned by a `Spectre.Turn.Handler`.
  - [Spectre.Turn.Handler.Request](Spectre.Turn.Handler.Request.md): Loaded request presented to an optional `Spectre.Turn.Handler`.

- Lifecycle and state
  - [Spectre.Awaitable](Spectre.Awaitable.md): Runtime state waiting for user input.
  - [Spectre.Effect](Spectre.Effect.md): Runtime or host work represented as data.
  - [Spectre.Instance.Lifecycle](Spectre.Instance.Lifecycle.md): Independent admission, authority, retention and activation axes for a Definition.
  - [Spectre.Lifecycle](Spectre.Lifecycle.md): Pure lifecycle state machine for effects and policy awaitables.
  - [Spectre.Policy](Spectre.Policy.md): Policy gate evaluator for pending effects.
  - [Spectre.Policy.Matcher](Spectre.Policy.Matcher.md): Pure matcher for policy accept and reject branches.
  - [Spectre.Policy.Resolution](Spectre.Policy.Resolution.md): Validated policy matcher decision shared by user text and trusted hosts.
  - [Spectre.Result](Spectre.Result.md): Output of a handled turn.
  - [Spectre.Skill.StateBinding](Spectre.Skill.StateBinding.md): Portable, generation-fenced state owned by one stable Skill and Definition.
  - [Spectre.State](Spectre.State.md): Conversation state owned by Spectre.
  - [Spectre.State.Codec](Spectre.State.Codec.md): Strict, versioned serialization for durable Spectre state.
  - [Spectre.State.Store](Spectre.State.Store.md): Durable state-adapter contract with optimistic concurrency.
  - [Spectre.Transition](Spectre.Transition.md): Immutable result of one validated lifecycle transition.

- Governance
  - [Spectre.Gate.Receipt](Spectre.Gate.Receipt.md): Content-addressed evidence emitted by one trusted governance checker.
  - [Spectre.Gate.Receipt.Ref](Spectre.Gate.Receipt.Ref.md): Content-addressed identity of a governance gate receipt.
  - [Spectre.Governance.Approval](Spectre.Governance.Approval.md): Separate host approval commit for a reviewed Candidate.
  - [Spectre.Governance.Approval.Policy](Spectre.Governance.Approval.Policy.md): Host-owned approval policy keyed by Candidate risk.
  - [Spectre.Governance.CandidateState](Spectre.Governance.CandidateState.md): Immutable governance state attached to a Definition Candidate.
  - [Spectre.Governance.ChangeSet](Spectre.Governance.ChangeSet.md): Immutable, non-executable proposal to derive a Definition from an observed
Instance Activation.
  - [Spectre.Governance.ChangeSet.Handler](Spectre.Governance.ChangeSet.Handler.md): Trusted implementation boundary for registered ChangeSet operations.
  - [Spectre.Governance.ChangeSet.Operation](Spectre.Governance.ChangeSet.Operation.md): One inert operation in a governed Definition ChangeSet.
  - [Spectre.Governance.ChangeSet.Registry](Spectre.Governance.ChangeSet.Registry.md): Host-owned registry mapping inert ChangeSet operation types to reviewed code.
  - [Spectre.Governance.Composer](Spectre.Governance.Composer.md): Trusted host composer for governed Definition Candidates.
  - [Spectre.Governance.Composition](Spectre.Governance.Composition.md): Accumulates the typed result of applying governed ChangeSet operations.
  - [Spectre.Governance.EvaluationDelta](Spectre.Governance.EvaluationDelta.md): Deterministic comparison of parent and Candidate results on the same corpus.
  - [Spectre.Governance.GC](Spectre.Governance.GC.md): Conservative administered GC planner for immutable governance artifacts.
  - [Spectre.Governance.GC.Plan](Spectre.Governance.GC.Plan.md): Content-addressed conservative garbage-collection decision.
  - [Spectre.Governance.Review](Spectre.Governance.Review.md): Host review commit for a composed Candidate.
  - [Spectre.Morph](Spectre.Morph.md): Host-facing facade for governed, declarative Agent evolution.
  - [Spectre.Morph.Change](Spectre.Morph.Change.md): Transient pipeline value for one governed Agent change.
  - [Spectre.Morph.DSL](Spectre.Morph.DSL.md): Declares the closed proposal surface of a compiled Agent.
  - [Spectre.Morph.Surface](Spectre.Morph.Surface.md): Canonical, closed declaration of the changes an Agent is willing to review.
  - [Spectre.Projection.HumanReport](Spectre.Projection.HumanReport.md): Mechanical human-review projection for one governed Definition change.

- Reflection and Forge
  - [Spectre.Experience](Spectre.Experience.md): Opt-in boundary for recording and querying redacted operational evidence.
  - [Spectre.Experience.Evidence](Spectre.Experience.Evidence.md): Immutable, redacted and non-canonical operational evidence.
  - [Spectre.Experience.Evidence.Ref](Spectre.Experience.Evidence.Ref.md): Content-addressed identity of redacted Experience evidence.
  - [Spectre.Experience.Redactor](Spectre.Experience.Redactor.md): Mechanical redaction boundary for opt-in Experience evidence.
  - [Spectre.Experience.Store](Spectre.Experience.Store.md): Opt-in persistence boundary for redacted Experience evidence.
  - [Spectre.Experience.Store.Memory](Spectre.Experience.Store.Memory.md): Volatile reference implementation of `Spectre.Experience.Store`.
  - [Spectre.Forge](Spectre.Forge.md): Data-only proposal plane for governed Definition changes.
  - [Spectre.Forge.Critic](Spectre.Forge.Critic.md): Compiled adapter contract for model-backed Forge critique, including Prism.
  - [Spectre.Forge.Critique](Spectre.Forge.Critique.md): Immutable result of one compiled Forge critic adapter.
  - [Spectre.Forge.OracleApproval](Spectre.Forge.OracleApproval.md): Independent host approval binding one proposed evaluation case to an oracle.
  - [Spectre.Forge.Proposal](Spectre.Forge.Proposal.md): Content-addressed Forge output wrapping an inert governance ChangeSet.
  - [Spectre.Reflection](Spectre.Reflection.md): Policy-gated host boundary for inspecting the active canonical Definition.
  - [Spectre.Reflection.Operation](Spectre.Reflection.Operation.md): Compiled, policy-gated operation executor for Reflection.
  - [Spectre.Reflection.Policy](Spectre.Reflection.Policy.md): Host-owned authorization policy for the Reflection operation.

- Routing
  - [Spectre.Router](Spectre.Router.md): Routes inbound input to a compiled Spectre rule.
  - [Spectre.Router.Arbitration](Spectre.Router.Arbitration.md): Input passed to a router arbitrator.
  - [Spectre.Router.Arbitrator](Spectre.Router.Arbitrator.md): Behaviour for final route selection from collected routing evidence.
  - [Spectre.Router.Candidate](Spectre.Router.Candidate.md): A single routing signal produced by a router evidence provider.
  - [Spectre.Router.Context](Spectre.Router.Context.md): Shared data passed through a Spectre router pipeline.
  - [Spectre.Router.Receipt](Spectre.Router.Receipt.md): Privacy-safe summary of one router-only evaluation.
  - [Spectre.Router.SemanticCache](Spectre.Router.SemanticCache.md): Adapter boundary for semantic cache lookups.
  - [Spectre.Router.SemanticCache.Learned](Spectre.Router.SemanticCache.Learned.md): Built-in semantic cache backed by static dataset rows and online examples.
  - [Spectre.Router.SemanticCache.Owner](Spectre.Router.SemanticCache.Owner.md): Stable owner for the built-in semantic-cache ETS tables and collection
lifecycle operations.

- Prompts and providers
  - [Spectre.LLM](Spectre.LLM.md): Small LLM boundary used by `Spectre.Runner`.
  - [Spectre.Prompt](Spectre.Prompt.md): Prompt resolver, renderer, and typed composition boundary.
  - [Spectre.Prompt.Fragment](Spectre.Prompt.Fragment.md): Resolved or canonical prompt content with its composition metadata.
  - [Spectre.Prompt.Operation](Spectre.Prompt.Operation.md): One typed prompt-composition declaration compiled from `inject`.

  - [Spectre.Prompt.Plan](Spectre.Prompt.Plan.md): Pure, typed prompt composition plan.
  - [Spectre.Provider.Call](Spectre.Provider.Call.md): Isolated timeout and crash boundary for external provider calls.
  - [Spectre.Provider.Failure](Spectre.Provider.Failure.md): Sanitized infrastructure failure returned by a provider call boundary.

- Actions
  - [Spectre.Action](Spectre.Action.md): Provider-neutral description of one action.
  - [Spectre.Action.Planner](Spectre.Action.Planner.md): Port implemented by optional action-planning libraries.
  - [Spectre.Action.Provider](Spectre.Action.Provider.md): Contract and safe invocation facade for action providers.
  - [Spectre.Action.Provider.Local](Spectre.Action.Provider.Local.md): Built-in provider for ordinary Elixir action modules.
  - [Spectre.Action.Spec](Spectre.Action.Spec.md): Planner-facing description of an action exposed by a provider.
  - [Spectre.ActionConfig](Spectre.ActionConfig.md): Resolves provider-neutral action configuration from an Agent definition.
  - [Spectre.ActionDispatcher](Spectre.ActionDispatcher.md): Capability boundary for invoking one provider-neutral action effect.
  - [Spectre.ActionExecutor](Spectre.ActionExecutor.md): Compatibility facade for the canonical `Spectre.Execution` workflow.
  - [Spectre.ActionHooks](Spectre.ActionHooks.md): Runs lifecycle hooks for executed actions.
  - [Spectre.ActionPlanner](Spectre.ActionPlanner.md): Provider-neutral facade over an optional action planner.
  - [Spectre.ActionProtection](Spectre.ActionProtection.md): Matches action effects against agent `protect` declarations.
  - [Spectre.Execution](Spectre.Execution.md): Pure-lifecycle execution workflow for action and extension-owned effects.

- Operational loops
  - [Spectre.Execution.Expression](Spectre.Execution.Expression.md): Closed expression language used by data-driven execution programs.
  - [Spectre.Execution.Handoff](Spectre.Execution.Handoff.md): Typed Flow/Work exchange carried as portable data.
  - [Spectre.Execution.Materialization](Spectre.Execution.Materialization.md): Exact, verified boundary between Skill routing and data-driven execution.
  - [Spectre.Execution.Materializer](Spectre.Execution.Materializer.md): Resolves a runtime Skill Work route into an exact execution materialization.
  - [Spectre.Execution.Migration](Spectre.Execution.Migration.md): Prepare/commit protocol for registered, pure Work state migrations.
  - [Spectre.Execution.Migration.Receipt](Spectre.Execution.Migration.Receipt.md): Integrity receipt for one committed registered state migration.
  - [Spectre.Execution.Program](Spectre.Execution.Program.md): Immutable, data-only program for a precise Work.
  - [Spectre.Execution.Rehearsal](Spectre.Execution.Rehearsal.md): Deterministic replay of a data-driven Work from recorded operation results.
  - [Spectre.Execution.Rehearsal.Report](Spectre.Execution.Rehearsal.Report.md): Privacy-safe receipt for a data-driven execution rehearsal.
  - [Spectre.Execution.Runtime](Spectre.Execution.Runtime.md): Admission boundary for a verified data-driven execution materialization.
  - [Spectre.Execution.Work](Spectre.Execution.Work.md): Compile-time DSL for declarative data-driven Work programs.
  - [Spectre.Instance.CheckpointStore](Spectre.Instance.CheckpointStore.md): Durable adapter boundary for the complete canonical Agent checkpoint.
  - [Spectre.Operation.Artifact](Spectre.Operation.Artifact.md): Portable reference to output produced by an operational loop.
  - [Spectre.Operation.Attempt](Spectre.Operation.Attempt.md): Revision- and epoch-fenced identity for one temporary Runner.
  - [Spectre.Operation.Budget](Spectre.Operation.Budget.md): Portable limits and consumption counters for one operational loop.
  - [Spectre.Operation.Control](Spectre.Operation.Control.md): Portable control plane for pause, update, resume and terminal stop.
  - [Spectre.Operation.Control.Command](Spectre.Operation.Control.Command.md): Durable, idempotent command for controlling one operational loop.
  - [Spectre.Operation.Controller](Spectre.Operation.Controller.md): Behaviour shared by Work, Vigil and library-owned operational controllers.
  - [Spectre.Operation.Definition](Spectre.Operation.Definition.md): Versioned definition loaded from a Work, Vigil or external controller.
  - [Spectre.Operation.Delivery](Spectre.Operation.Delivery.md): Authorizes proactive delivery without performing transport I/O.
  - [Spectre.Operation.Delivery.Consent](Spectre.Operation.Delivery.Consent.md): Revocable, expiring consent for proactive delivery to one destination.
  - [Spectre.Operation.Delivery.Policy](Spectre.Operation.Delivery.Policy.md): Deterministic consent, deduplication, rate and quiet-hours policy.
  - [Spectre.Operation.Delivery.Receipt](Spectre.Operation.Delivery.Receipt.md): Portable audit receipt for one governed delivery decision.
  - [Spectre.Operation.Event](Spectre.Operation.Event.md): Committed event envelope shared by Work, Vigil and external controllers.
  - [Spectre.Operation.Events](Spectre.Operation.Events.md): Local subscription boundary for already-committed operational events.
  - [Spectre.Operation.Execution](Spectre.Operation.Execution.md): Portable output envelope produced by an operational executor.
  - [Spectre.Operation.ExecutionContext](Spectre.Operation.ExecutionContext.md): Ephemeral context passed to one registered operation executor.
  - [Spectre.Operation.Executor](Spectre.Operation.Executor.md): Executes one closed-catalog operation inside a temporary Runner.
  - [Spectre.Operation.Loop](Spectre.Operation.Loop.md): Canonical portable state of one operational loop.
  - [Spectre.Operation.Memory](Spectre.Operation.Memory.md): Selective post-commit port for Mnemonic or an application memory adapter.
  - [Spectre.Operation.Monitor](Spectre.Operation.Monitor.md): Classifies Runner termination without deciding retry outside the Agent.
  - [Spectre.Operation.Outcome](Spectre.Operation.Outcome.md): Typed terminal outcome of an operational loop.
  - [Spectre.Operation.Policy](Spectre.Operation.Policy.md): Deterministic authorization gate for a registered operational capability.
  - [Spectre.Operation.Progress](Spectre.Operation.Progress.md): Revision-fenced, rate-limited progress emitted by a temporary Runner.
  - [Spectre.Operation.Ref](Spectre.Operation.Ref.md): Stable, portable reference to a Work, Vigil or external controller loop.
  - [Spectre.Operation.Registry](Spectre.Operation.Registry.md): Resolves the closed operation catalog visible to one loop step.
  - [Spectre.Operation.Request](Spectre.Operation.Request.md): One portable request for a registered operation.
  - [Spectre.Operation.Result](Spectre.Operation.Result.md): Correlated terminal result of exactly one operational attempt.
  - [Spectre.Operation.Retry](Spectre.Operation.Retry.md): Portable retry and backoff policy for a registered operation.
  - [Spectre.Operation.Runner](Spectre.Operation.Runner.md): Temporary supervised process for one operation and one fenced attempt.
  - [Spectre.Operation.RunnerSupervisor](Spectre.Operation.RunnerSupervisor.md): Dynamic supervisor for isolated one-attempt operational Runners.
  - [Spectre.Operation.Runtime](Spectre.Operation.Runtime.md): Pure transition engine for Work, Vigil and external operational controllers.
  - [Spectre.Operation.Spec](Spectre.Operation.Spec.md): Immutable registry entry for one operation available to an operational loop.
  - [Spectre.Operation.Update](Spectre.Operation.Update.md): One committed update to a loop's effective context.
  - [Spectre.Operation.Validator](Spectre.Operation.Validator.md): Safe validation boundary for registered operation input, output and updates.
  - [Spectre.Operation.View](Spectre.Operation.View.md): Privacy-safe, read-only projection of a committed operational loop.
  - [Spectre.Operation.Wait](Spectre.Operation.Wait.md): Portable wait boundary retained by an operational loop without a live Runner.
  - [Spectre.Projection.Execution](Spectre.Projection.Execution.md): Deterministic, privacy-safe view of one data-driven Work materialization.
  - [Spectre.Prompt.Materializer](Spectre.Prompt.Materializer.md): Closed renderer for canonical prompt fragments.
  - [Spectre.Prompt.Receipt](Spectre.Prompt.Receipt.md): Privacy-safe evidence for one effective governed prompt fragment.
  - [Spectre.Vigil](Spectre.Vigil.md): Versioned durable observation loop built on the shared operational runtime.
  - [Spectre.Work](Spectre.Work.md): Versioned definition for a precise, asynchronous operational procedure.

- Extensions
  - [Spectre.Extension](Spectre.Extension.md): Compile-time contract for on-demand Spectre libraries.
  - [Spectre.Stack.Contract.V1](Spectre.Stack.Contract.V1.md): Executable version-one conformance contract for Stack packages.
  - [Spectre.Stack.Contract.V2](Spectre.Stack.Contract.V2.md): Mandatory runtime Stack contract for reflective Definition Manifests.
  - [Spectre.Stack.Definition](Spectre.Stack.Definition.md): Compiled, immutable environment installed for an Agent.
  - [Spectre.Stack.Installable](Spectre.Stack.Installable.md): Versioned contract implemented by packages that can be installed in a Stack.
  - [Spectre.Stack.Installation](Spectre.Stack.Installation.md): Immutable result of compiling one package inside a Stack.
  - [Spectre.Stack.Package](Spectre.Stack.Package.md): Versioned, immutable manifest published by a Stack-installable package.
  - [Spectre.Stack.Ref](Spectre.Stack.Ref.md): Logical reference to one capability or runtime resource in a resolved Stack.
  - [Spectre.Stack.Runtime](Spectre.Stack.Runtime.md): Caller-owned supervisor for resources declared by a resolved Stack.

- Classifiers and evaluation
  - [Spectre.Classifier](Spectre.Classifier.md): Runtime for Spectre's lightweight local vector classifier.
  - [Spectre.Classifier.Embedding](Spectre.Classifier.Embedding.md): Behaviour for classifier embedding adapters.
  - [Spectre.Classifier.Encoder](Spectre.Classifier.Encoder.md): Configurable embedding boundary for Spectre classifier artifacts.
  - [Spectre.Classifier.Trainer](Spectre.Classifier.Trainer.md): Builds a lightweight vector classifier artifact for Spectre routing.
  - [Spectre.Eval](Spectre.Eval.md): Route-only routing regression evaluation.
  - [Spectre.Eval.Case](Spectre.Eval.Case.md): One version-controlled routing expectation.
  - [Spectre.Eval.Report](Spectre.Eval.Report.md): Aggregate routing quality and LLM-use metrics for an evaluation corpus.

  - [Spectre.Eval.Result](Spectre.Eval.Result.md): Result of comparing one routing receipt with its evaluation expectations.

  - [Spectre.Training.Dataset](Spectre.Training.Dataset.md): Builds classifier datasets from Spectre agent routing metadata.

- Journal and operations
  - [Spectre.Journal](Spectre.Journal.md): Privacy-safe public journal boundary for optional Spectre extensions.
  - [Spectre.Journal.Buffer](Spectre.Journal.Buffer.md): Bounded asynchronous delivery queue for observational journal writes.
  - [Spectre.Journal.Record](Spectre.Journal.Record.md): Versioned, privacy-aware explanation record for one runtime decision.
  - [Spectre.Journal.Recorder](Spectre.Journal.Recorder.md): Builds, filters, and delivers observational journal records.
  - [Spectre.Journal.Store](Spectre.Journal.Store.md): Storage boundary for structured Spectre journal records.
  - [Spectre.Monitor](Spectre.Monitor.md): Failure boundary for host delivery around a Spectre agent.
  - [Spectre.Telemetry](Spectre.Telemetry.md): Privacy-safe operational telemetry boundary.

- Supporting API
  - [Spectre.Action.Provider.Mount](Spectre.Action.Provider.Mount.md): Trusted binding between a provider identifier and its implementation module.
  - [Spectre.ActionGuards](Spectre.ActionGuards.md): Runs `before_action` guards right before an action capability is invoked.
  - [Spectre.Classifier.Embeddings.ExFastembed](Spectre.Classifier.Embeddings.ExFastembed.md): Default classifier embedding adapter backed by `ExFastembed`.

  - [Spectre.Classifier.Math](Spectre.Classifier.Math.md): Small vector math helpers used by classifier and embedding strategies.
  - [Spectre.Context](Spectre.Context.md): Per-turn runtime context.
  - [Spectre.Definition](Spectre.Definition.md): Compiled behavioral description shared by Agents and Skills.
  - [Spectre.Definition.Validator](Spectre.Definition.Validator.md): Compile-time validation for Agent definitions and mounted Skills.

  - [Spectre.Effect.Executor](Spectre.Effect.Executor.md): Safe capability port for extension-owned, non-action effects.
  - [Spectre.Effect.Executor.Mount](Spectre.Effect.Executor.Mount.md): Immutable registration of one executor for a non-action effect kind.

  - [Spectre.EffectConfig](Spectre.EffectConfig.md): Resolves extension-contributed executors for non-action effect kinds.

  - [Spectre.Event.SchemaRegistry](Spectre.Event.SchemaRegistry.md): Host-owned compatibility gate for global Event payload schemas.
  - [Spectre.Extension.Mount](Spectre.Extension.Mount.md): Compile-time registration of one optional Spectre extension.
  - [Spectre.Flow.Constraint](Spectre.Flow.Constraint.md): Namespaced constraint contributed to a compiled flow by an extension.
  - [Spectre.Governance.Checker.Declarative](Spectre.Governance.Checker.Declarative.md): Model-free replay and regression checker for closed declarative Skill replies.
  - [Spectre.Governance.Checker.Declarative.Evaluator](Spectre.Governance.Checker.Declarative.Evaluator.md): Executes normalized evaluation cases against an exact loaded Definition.
  - [Spectre.Governance.Checker.Declarative.ReceiptIssuer](Spectre.Governance.Checker.Declarative.ReceiptIssuer.md): Issues replay and regression receipts for one declarative evaluation delta.
  - [Spectre.Governance.Checker.Declarative.Rehearsability](Spectre.Governance.Checker.Declarative.Rehearsability.md): Defines the closed live behavior that the declarative checker can replay.
  - [Spectre.Governance.Checker.Declarative.Shape](Spectre.Governance.Checker.Declarative.Shape.md): Provides one stable shape vocabulary for declarative-checker errors.

  - [Spectre.Governance.GC.InventoryProvider](Spectre.Governance.GC.InventoryProvider.md): Trusted host contract for a fenced, complete GC inventory snapshot.
  - [Spectre.Identity](Spectre.Identity.md): Generates durable identifiers shared by runtime entities.
  - [Spectre.Inference](Spectre.Inference.md): Neutral per-invocation cognitive selection and completion boundary.
  - [Spectre.Inference.Constraints](Spectre.Inference.Constraints.md): Hard limits and soft preferences attached to one inference request.
  - [Spectre.Inference.Profile](Spectre.Inference.Profile.md): Provider-neutral cognitive capability profile.

  - [Spectre.Inference.Request](Spectre.Inference.Request.md): One typed, purpose-labelled inference invocation.

  - [Spectre.Inference.Response](Spectre.Inference.Response.md): Normalized inference response with optional usage and provider metadata.

  - [Spectre.Inference.Selection](Spectre.Inference.Selection.md): Frozen model selection for one inference attempt.

  - [Spectre.Inference.Selector](Spectre.Inference.Selector.md): Behaviour implemented by inference-selection policies.

  - [Spectre.Inference.Selector.Default](Spectre.Inference.Selector.Default.md): Compatibility selector for Agents without a cognitive-routing extension.

  - [Spectre.Input](Spectre.Input.md): Normalized inbound user turn.
  - [Spectre.Input.Pipeline](Spectre.Input.Pipeline.md): Runs input plugs before the Spectre runtime handles a turn.
  - [Spectre.Input.Pipeline.Spec](Spectre.Input.Pipeline.Spec.md): Initialized input plug declaration.
  - [Spectre.Input.Plug](Spectre.Input.Plug.md): Behaviour for Spectre input normalization/enrichment stages.
  - [Spectre.Input.Plugs.NormalizeText](Spectre.Input.Plugs.NormalizeText.md): Normalizes `input.text` before Spectre handles a turn.
  - [Spectre.Input.Source](Spectre.Input.Source.md): Provider-neutral origin attached to a normalized Spectre input.
  - [Spectre.Invocation.Receipt](Spectre.Invocation.Receipt.md): Correlated internal receipt returned to an Agent Instance mailbox.
  - [Spectre.Morph.GovernancePipeline](Spectre.Morph.GovernancePipeline.md): Executes the durable governance commits behind the `Spectre.Morph` facade.
  - [Spectre.Morph.Options](Spectre.Morph.Options.md): Validates the closed keyword contracts exposed by the Morph facade.
  - [Spectre.Morph.SkillProposal](Spectre.Morph.SkillProposal.md): Builds the closed reply-only Skill proposals accepted by Morph.
  - [Spectre.Morph.StableName](Spectre.Morph.StableName.md): Normalizes existing stable identifiers without creating runtime atoms.
  - [Spectre.Morph.Surface.EvaluationObligations](Spectre.Morph.Surface.EvaluationObligations.md): Derives mandatory evaluation cases from an immutable Definition diff.
  - [Spectre.Morph.Surface.MountIndex](Spectre.Morph.Surface.MountIndex.md): Indexes canonical Skill mounts and derives their parent-to-candidate diff.
  - [Spectre.Morph.Surface.Mutation](Spectre.Morph.Surface.Mutation.md): Describes one immutable Skill-mount difference between two Agent Definitions.
  - [Spectre.Morph.Surface.Policy](Spectre.Morph.Surface.Policy.md): Applies the immutable Morph proposal ceiling to governance operations.
  - [Spectre.Operation.Runtime.Contract](Spectre.Operation.Runtime.Contract.md): Definition and declaration validation for the operation runtime.
  - [Spectre.Operation.Runtime.Controls](Spectre.Operation.Runtime.Controls.md): Control-plane command handling for the operation runtime.
  - [Spectre.Operation.Runtime.Recovery](Spectre.Operation.Runtime.Recovery.md): Restart recovery and checkpoint validation for the operation runtime.
  - [Spectre.Operation.Runtime.Results](Spectre.Operation.Runtime.Results.md): Result reduction and completion evaluation for the operation runtime.
  - [Spectre.Operation.Runtime.Support](Spectre.Operation.Runtime.Support.md): Shared vocabulary for the operation runtime modules.
  - [Spectre.Operation.Runtime.Transitions](Spectre.Operation.Runtime.Transitions.md): Pure loop-state transition builders for the operation runtime.
  - [Spectre.Pipeline](Spectre.Pipeline.md): Minimal Plug-style pipeline executor for Spectre router stages.
  - [Spectre.Pipeline.Spec](Spectre.Pipeline.Spec.md): Initialized router plug declaration.
  - [Spectre.Projection.Routing](Spectre.Projection.Routing.md): Deterministic, non-executable routing view of a canonical Definition.
  - [Spectre.Route](Spectre.Route.md): Normalized routing decision returned by Spectre.
  - [Spectre.Router.Arbitrators.Default](Spectre.Router.Arbitrators.Default.md): Default evidence arbitrator for Spectre routers.

  - [Spectre.Router.DefaultPipeline](Spectre.Router.DefaultPipeline.md): Default router pipeline for Spectre agents.
  - [Spectre.Router.IndexProfile](Spectre.Router.IndexProfile.md): Versioned identity for a rebuildable routing cache or index.
  - [Spectre.Router.LLMClassifier](Spectre.Router.LLMClassifier.md): One-label LLM fallback classifier for Spectre routes.

  - [Spectre.Router.LocalClassifier](Spectre.Router.LocalClassifier.md): Adapter boundary for local intent classifiers.
  - [Spectre.Router.Plug](Spectre.Router.Plug.md): Behaviour for Spectre router pipeline stages.
  - [Spectre.Router.Plugs.Arbitrate](Spectre.Router.Plugs.Arbitrate.md): Turns accumulated routing evidence into a route decision.
  - [Spectre.Router.Plugs.BagDistance](Spectre.Router.Plugs.BagDistance.md): Bag-distance evidence provider for short example matching.
  - [Spectre.Router.Plugs.EmbeddingSimilarity](Spectre.Router.Plugs.EmbeddingSimilarity.md): Embedding similarity evidence provider.
  - [Spectre.Router.Plugs.JaroDistance](Spectre.Router.Plugs.JaroDistance.md): Jaro-distance evidence provider for typo-tolerant short examples.
  - [Spectre.Router.Plugs.LLMFallback](Spectre.Router.Plugs.LLMFallback.md): Legacy LLM fallback classifier plug.
  - [Spectre.Router.Plugs.LocalClassifier](Spectre.Router.Plugs.LocalClassifier.md): Local classifier evidence provider.
  - [Spectre.Router.Plugs.Regex](Spectre.Router.Plugs.Regex.md): Deterministic regex evidence provider for the router pipeline.
  - [Spectre.Router.Plugs.SemanticCacheExact](Spectre.Router.Plugs.SemanticCacheExact.md): Exact semantic-cache lookup before classifier fallback.
  - [Spectre.Router.Plugs.SemanticCacheSearch](Spectre.Router.Plugs.SemanticCacheSearch.md): Semantic-cache search after local classifier evidence.
  - [Spectre.Router.Plugs.Terminalize](Spectre.Router.Plugs.Terminalize.md): Applies terminal-route metadata after arbitration.
  - [Spectre.Router.SemanticCache.Learned.Index](Spectre.Router.SemanticCache.Learned.Index.md): Vettore-backed semantic index and embedding acquisition for the learned
cache.
  - [Spectre.Router.SemanticCache.Learned.Online](Spectre.Router.SemanticCache.Learned.Online.md): ETS-backed store for online learned semantic-cache examples.
  - [Spectre.Router.SemanticCache.Learned.Rows](Spectre.Router.SemanticCache.Learned.Rows.md): Row and label vocabulary for the built-in learned semantic cache.
  - [Spectre.Router.SemanticCache.Learned.Snapshot](Spectre.Router.SemanticCache.Learned.Snapshot.md): Portable snapshot export and import for the learned semantic cache.
  - [Spectre.Router.SemanticCache.Learned.Sources](Spectre.Router.SemanticCache.Learned.Sources.md): Static row sources for the learned semantic cache.
  - [Spectre.Router.Support](Spectre.Router.Support.md): Shared router normalization and logging helpers.
  - [Spectre.Rule](Spectre.Rule.md): Declarative routing rule compiled from `Spectre.Agent` DSL.
  - [Spectre.Run.Codec](Spectre.Run.Codec.md): Serializes a `Spectre.Run` to and from its portable checkpoint binary.
  - [Spectre.Run.Value](Spectre.Run.Value.md): Portable value codec for run state.
  - [Spectre.Runner](Spectre.Runner.md): Executes routed handlers without crossing action boundaries accidentally.
  - [Spectre.Runtime.Persistence](Spectre.Runtime.Persistence.md): State and memory persistence boundaries for `Spectre.Runtime`.
  - [Spectre.Runtime.SkillDispatch](Spectre.Runtime.SkillDispatch.md): Dispatches runtime-authored Skills from the Definition pinned to a Run.
  - [Spectre.Skill.Applicability](Spectre.Skill.Applicability.md): Structured, portable applicability contract for one Skill.
  - [Spectre.Skill.Definition](Spectre.Skill.Definition.md): Unified view of compiled and runtime Skill Definitions.
  - [Spectre.Skill.Mount](Spectre.Skill.Mount.md): Compile-time mount of a reusable `Spectre.Skill` inside an Agent.
  - [Spectre.Skill.PromptBudget](Spectre.Skill.PromptBudget.md): Closed per-Skill prompt budget derived from canonical fragments.
  - [Spectre.Skill.Runtime](Spectre.Skill.Runtime.md): Immutable, host-governed registry for compiled and runtime Skills.
  - [Spectre.Skill.Runtime.Loader](Spectre.Skill.Runtime.Loader.md): Restores the governed Skill runtime for one exact Agent Definition Ref.
  - [Spectre.Skill.Runtime.Response](Spectre.Skill.Runtime.Response.md): Public result of one declarative runtime Skill response.
  - [Spectre.Stack.Binding](Spectre.Stack.Binding.md): Binds an immutable Stack installation to one Agent definition.
  - [Spectre.Stack.DSL](Spectre.Stack.DSL.md): Parser helper for package-local Stack declarations.
  - [Spectre.Stack.Value](Spectre.Stack.Value.md): Portability checks and identity helpers for stack entry values.
  - [Spectre.Turn.Decision](Spectre.Turn.Decision.md): Reduces a `%Spectre.Result{}` into the host's next lifecycle decision.

## Mix Tasks

- Mix tasks
  - [mix spectre.classifier.dataset](Mix.Tasks.Spectre.Classifier.Dataset.md): Exports classifier training rows from a Spectre agent.
  - [mix spectre.classifier.download_model](Mix.Tasks.Spectre.Classifier.DownloadModel.md): Downloads/loads the embedding model used by Spectre classifier artifacts.
  - [mix spectre.classifier.train](Mix.Tasks.Spectre.Classifier.Train.md): Trains Spectre's local classifier artifact from a JSON dataset.
  - [mix spectre.eval](Mix.Tasks.Spectre.Eval.md): Evaluates Spectre routing accuracy and LLM usage from a JSONL corpus.

