leonard_agent 0.3.1

Web-compatible harness library.

Changelog

0.3.1

• Fix: schema-declared integer and number tool arguments now normalize lossless quoted SwiftInfer numerics before strict ActionSchema validation. Outgoing schemas, ActionValidator, and retry budgets remain strict.

0.3.0

Promotes 0.3.0-rc.1 to stable. The candidate's full change list is under that entry below; the break is restated here because this is the version a consumer on 0.2.x upgrades to.

• Breaking: owning VM-service connections live in leonard_agent_io.dart. The static connect constructors — VmServiceClient.connect, LeonardSession.connect and MultiHostSession.connectAll — are gone from the default, web-safe library. Migration: import package:leonard_agent/leonard_agent_io.dart and call the top-level connectVmServiceClient, connectLeonardSession or connectMultiHostSession. A consumer that only wraps an already-connected VmService (VmServiceClient.fromVmService) is unaffected.
• Breaking: SwiftInferChatOptions's sampling fields are nullable. maxTokens, temperature, topP, topK, presencePenalty and repetitionPenalty moved from int/double to int?/double?, so an unset override is omitted from the request instead of being sent as a default. Migration: null-check before reading one.
• Fix: device tool descriptors carried in a handshake are parsed into agent manifest entries and core descriptors route through coreTools, so validation rejects a malformed call before device dispatch. A legacy names-only handshake keeps the permissive projection (lenny#123).
• Add: the loop driver and conversation builder carry the trim-proof scratchpad, so a remembered result survives history trimming and is read back by recall (lenny#116).

0.3.0-rc.1

• Breaking: owning VM-service connections moved to leonard_agent_io.dart. The static connect constructors — VmServiceClient.connect, LeonardSession.connect and MultiHostSession.connectAll — are gone from the default, web-safe library. Migration: import package:leonard_agent/leonard_agent_io.dart and call the top-level connectVmServiceClient, connectLeonardSession or connectMultiHostSession. A consumer that only wraps an already-connected VmService (VmServiceClient.fromVmService) is unaffected and stays on leonard_agent.dart.
• Fix: model-response transport faults are classified as retryable failed turns and their diagnostic context carries into the terminal agent-stuck footer. Any exception still escaping the session loop is named and scrubbed, so a harness error is never anonymous.
• Fix: prior reasoning replays as native thinking blocks in assistant history, so swift-infer populates reasoning content without template-specific marker parsing.
• Fix: unset sampling overrides are omitted rather than sent; shared driver token and effort defaults are added, with operator overrides.
• Fix: the swift-infer qwen tier advertises its native context window, and the dogfood harness reads the shared capability registry.
• Fix: tooltip-only semantics are promoted to labels, distinct tooltips are preserved as hints, and the optional field carries through typed observations, so icon buttons are addressable by drivers.
• Fix: self-drive completion is gated on observed timeline evidence — a scenario-declared node pattern must match and captured row tokens are cross-checked against the completion reason — and diagnostic evidence is preserved when an observation fails.
• Raise the leonard_contract floor to ^0.2.2; the declared ^0.2.0 floor was unsatisfiable against this package's own genesis_perception range.

0.2.0

• Breaking: VM-service methods now use ext.leonard.*. Construct names with kLeonardExtensionPrefix from leonard_contract.
• The VM-service client now consumes the shared Leonard wire contract.

0.1.6

• Provider response metadata is now persisted with the trajectory turn. The served model id and the streamed result identifier carry through model decisions into trajectory turns, so swift-infer observability stays on dartantic's standard result-identifier seam.
• New ModelDecision.modelMetadata (optional, defaults to empty). Dartantic-backed providers populate served_model_id and an explicit provider_request_id; custom providers may leave it empty.

0.1.5

• Perception nodes now carry identifier and value end-to-end to the brain. SemanticsNode parses and re-emits the stable, locale-independent identifier (from Semantics(identifier:)) and the node's value (text-field contents / secure-field bullets); both join ==/hashCode, so a field filling in shows up in the diff. The host emitted these already — the agent model had been dropping them, so the brain only saw them on the native channel.
• The bundled agent guide (kDefaultAgentsMd) documents the split: read label to understand what a node is, use identifier as the stable handle for addressing it across locales/sessions, and always act by integer node_id.

0.1.4

• Multi-host attach: MultiHostSession attaches to N VM-service hosts at once, merges each host's perception fragment into one observation (side-by-side, keyed by namespace), and routes each tool call to the owning host by namespace (core.* → the Flutter host, native.* → the native channel). A new SessionSurface interface is implemented by BOTH the unchanged single-host LeonardSession and MultiHostSession, so the loop drives either transparently. The agent context-switches by perception, not by hardcoded mode flags.

0.1.3

• HandshakeResult gains a capabilities field: host-level features that are reachable but are NOT namespaced tools (so they never appear under extensions) — notably screenshot. The handshake parse reads the new capabilities array and is tolerant of its absence (older bindings parse to an empty list). Lets a driver list screenshot where agents look instead of concluding "no such capability" from the tool manifest alone.

0.1.2

• Adopt dartantic as the model-backend seam: a single DartanticModelProvider drives any backend — swift-infer via lenny's custom ChatModel, Anthropic and OpenAI via stock dartantic models. The hand-rolled per-provider classes are removed; the loop keeps retry ownership and the SchemaRejection contract.
• Anthropic backend defaults are now compatible with extended thinking: a non-forcing tool_choice (auto) and no temperature override. Anthropic rejects a forcing tool_choice or any non-1 temperature while thinking is enabled, so the previous defaults returned request-time 400s when driving Claude. Thinking stays on; the driver's retry covers a rare prose-only turn.
• Fix (Anthropic): per-turn observation context is no longer dropped. The dartantic Anthropic mapper serializes only the tool_result block of a tool-bearing user message and discards sibling text parts, so from turn 1 on the model never saw the observation — it was driven blind after the first turn (the swift-infer path is unaffected; its mapper keeps the text). The Anthropic backend now folds the observation + diff into the tool_result body so the model sees the live screen every turn.
• Observation: expose scroll extent on scrollable nodes.

0.1.1

• Fix: bound runaway model output. The swift-infer provider now aborts a response once it streams a large amount of reasoning text with no tool call in sight, surfacing a retryable SchemaRejection instead of letting weaker models ruminate all the way to max_tokens — the "endless stream, no tool call" failure. The loop retries with a fresh sample.

0.1.0

Initial release.