leonard_devtools 0.4.0-dev.1

DevTools extension for Leonard.

Changelog

0.4.0-dev.1

• Breaking: ProviderConfig no longer has baseUrl or headersFor; they moved to the new HttpProviderConfig (swift-infer, anthropic, openai), since the ACP provider has no HTTP endpoint. ModelCatalog.fetch now takes an HttpProviderConfig. Migration: match HttpProviderConfig before reading baseUrl/headersFor or calling fetch; AcpUiConfig resolves its models through the Dart Tooling Daemon.
• Add model-provider sessions over the Dart Tooling Daemon: the prompt panel can select an ACP harness and its models.
• Fix: on a panel with no stored provider config, the Model picker now loads from the default swift-infer config the settings form shows. It stayed empty while Test connection reported OK, and Start silently did nothing.
• Fix: the Timeline replays the session's records to a listener that attaches late, so opening the tab after a run no longer shows "Waiting for trajectory records..." for the whole session.
• The panel's "Wall-clock budget" and "Max turns" fields now take effect (requires leonard_agent ^0.3.2-dev.1, which also stops the panel listing core tools twice).

0.3.0

Promotes 0.3.0-rc.1 to stable. The candidate's full change list is under that entry below.

• Breaking: requires leonard_agent ^0.3.0. Migration: move the leonard_agent constraint to ^0.3.0. This package consumes SwiftInferReasoningEffort, defaultSwiftInferOptions and SwiftInferChatOptions.reasoningEffort, which exist only from that version.
• Fix: the prompt panel carries the handshake's device tool descriptors into the driver's tool set (lenny#123).

0.3.0-rc.1

• Require leonard_agent ^0.3.0-rc.1. This package uses the borrowed VmServiceClient.fromVmService path, which is unchanged, but it also consumes SwiftInferReasoningEffort, defaultSwiftInferOptions and SwiftInferChatOptions.reasoningEffort, which exist only from that version.
• Fix: the panel retains durable self-drive session markers. A monotonic generation is persisted beside terminal run status and the pair renders in the visible status chip, so an inner run that completes between two observations stays detectable.
• Fix: the browser-only direct OpenAI provider choice is disabled, with a proxy-oriented explanation in its place; persisted proxy configurations remain editable.
• Raise the leonard_contract floor to ^0.2.2.

0.2.1

• feat: diagnostics tree inspector panel. The single shipped extension now presents Conversation and Diagnostics modes; Diagnostics renders Genesis diagnostics contract 1, loaded on demand over the existing DevTools VM connection via ext.leonard.core.get_diagnostics_tree — only while the panel is open. Apps need only a plain leonard_flutter dependency; no adopter dependency on leonard_devtools or genesis_foundation is required.

0.2.0

• Breaking: the DevTools extension probes and consumes ext.leonard.* extensions (protocol version 2); requires an app on leonard_agent 0.2.0.

0.1.1

• Provider construction moves to the DartanticModelProvider seam (the agent's unified backend factory).
• Fix: a run-level provider/HTTP failure no longer crashes the panel. The session run future now surfaces a terminal error status instead of escaping its unawaited wrapper as an unhandled async error (which previously took down the whole DevTools extension when, e.g., the model request threw).

0.1.0

Initial release.