Module contracts

Available on crate feature computer-use only.

Expand description

camelCase wire contracts for computer-use-mcp.

These types round-trip the JSON payloads shared with the TypeScript runtime and enforce disclosure-safe invariants on deserialization (digest-only postconditions, value-free sensitivity evidence, bounded approval scopes). They carry no desktop-actuation logic — the computer-use-mcp server remains authoritative for policy, identity, leases, and idempotency.

The submodules group related contracts:

action — action classification, resource context, provenance, postconditions, and the ActionEnvelope
target — target evidence and value-free sensitivity signals
approval — disclosure-safe ApprovalGrant scopes
receipt — capabilities, policy decisions, previews, and receipts
lease — one-writer control leases and target reservations
session — session lifecycle, events, follow-ups, and deletion results
safety — the versioned cross-runtime safety corpus

Modules§

action: Action classification, resource context, provenance, postconditions, and the immutable ActionEnvelope proposed by the graph and enforced by the runtime.
approval: Disclosure-safe approval grants returned by the runtime approval boundary.
lease: One-writer control leases and non-authoritative target reservations.
receipt: Capability, policy, preview, and execution-receipt contracts.
safety: Versioned cross-runtime deterministic safety corpus shared with the TypeScript fake-desktop harness.
session: Session lifecycle, event, follow-up, and deletion contracts.
target: Target evidence and value-free accessibility sensitivity contracts.

Structs§

ActionEnvelope: Immutable action proposed by the graph and enforced by the runtime.
ActionPreview: Result of preview_action used for deterministic graph routing.
ActionProvenance: Provenance describing whether an action derives from untrusted instructions.
ActionResourceContext: Optional resource identifiers describing what an action touches.
ApprovalGrant: Read-only, disclosure-safe authority returned by the runtime approval boundary.
Bounds: Axis-aligned bounding box in display coordinates.
ControlLease: One-writer desktop control lease.
ExecutionCapability: A verified execution capability advertised by the runtime.
ExecutionReceipt: Idempotency receipt returned by execute_action.
LeaseBoundaries: App/window/display boundaries restricting a ControlLease.
PolicyDecision: A policy decision for a previewed action.
PostconditionEvidence: A single postcondition and whether it was satisfied.
RuntimeSession: A runtime session and its current lifecycle state.
SafetyCorpus: Versioned cross-runtime deterministic safety corpus.
SafetyExpectation: Expected effects, restores, and status for a SafetyScenario.
SafetyScenario: A single deterministic safety scenario and its expected outcome.
SessionCompletionEvidence: Evidence describing how a session reached completion.
SessionDeletionResult: Principal-bound result of deleting one terminal runtime session’s durable data.
SessionEvent: Stable supervisor event used for ADK/runtime trace correlation.
SessionFollowUp: Principal-owned, monotonic steering instruction consumed by an ADK graph.
SessionFollowUpPage: A page of follow-ups plus the cursor for the next page.
TargetEvidence: Evidence binding an action to a fresh desktop observation.
TargetReservation: Non-authoritative, expiring planner intent for multi-agent conflict checks.
TargetReservationScope: Application/window scope of a TargetReservation.
TargetSensitivityEvidence: Value-free native accessibility evidence used for action risk and revalidation.

Enums§

ActionClass: Operation-aware action class used by authorization and policy.
ActionPostcondition: Digest-only expected state independently verified by computer-use-mcp.
ApprovalGrantScope: Scope of an ApprovalGrant.
ExecutionMode: Enforced execution mode selected for an action.
ReceiptStatus: Terminal status of an execution receipt.
TargetSensitivityAssessment: Conclusion of an accessibility-based target sensitivity check.
TargetSensitivitySignal: A single value-free signal contributing to a sensitivity assessment.
TargetSensitivitySource: Source of a TargetSensitivityEvidence assessment.