Module capability

Structs§

Budget: Per-session cost budget enforced by the supervisor.
CapabilityConstraints: Constraints the model declares on the tool it is requesting. All fields default to the most restrictive value if absent.
CapabilityManifest: Full manifest created by the tool architect (Phase 2+). In Phase 1 this is a data type only — no generation logic exists yet.
CapabilityMemoryRecord: Capability memory entry — fingerprint stored after a run. The binary is destroyed; only the pattern, constraints, and metrics survive.
CapabilityRequest: A structured request the model emits when text reasoning is genuinely insufficient for a computational task. The model NEVER generates or runs code — it only describes what it needs.
ModelProposalOutput: Wraps the model’s full propose-stage output. The telemetry fields are at the top level (flattened) so existing model responses without capability_request still parse unchanged.
Permissions: Fine-grained permission set attached to a generated tool manifest.
PolicyViolation: A policy rule that was violated. Returned as a list from policy::check_request.
ResourceLimits: Hard resource limits enforced by the sandbox runtime.
ToolMetrics: Measured execution metrics from one tool run (mock or real).
ToolRunReport: Full result returned by the supervisor after processing one CapabilityRequest.

VerificationKind: Verification steps the supervisor must run before a tool may execute.