v0.7.0

What's Changed

• Added a VS Code extension for Contract4Agents syntax highlighting.
• Added syntax highlighting support for Contract4Agents eval files.
• Improved highlighting for unquoted list items, mixed prose lists, expressions, guards, assertions, fixtures, trace calls, and dotted tool/function references.
• Added VS Code extension documentation and release workflow support.
• Improved docs validation by ignoring generated, dependency, cache, and build directories such as node_modules, dist, and .contract.
• Cleaned up project dependencies and stale decision documentation.

v0.6.0

What's Changed

• Added drift checks for per-agent hosted tool configurations in capability registries.
• Hosted tool drift diagnostics now report mismatches for the specific agent whose effective config differs from the contract declaration.
• Added validation for hosted_tools.*.agent_configs, including non-empty agent names and string-only config values.
• Added a new CAP090 diagnostic for stale hosted tool agent_configs entries that no longer match any contract declaration.

v0.5.0

What's Changed

• Added run spec derived value declarations, allowing contracts to declare expected derived inputs and validate that supplied values are present and correctly typed.
• Added derived value invariant checks for run specs, including set-style relations such as subset, contains-all, equality, intersection, and disjointness.
• Extended conditional run spec assertions so trace conditions can require either trace expectations or derived value data relations.
• Improved assertion failures for missing, unknown, unsupported, or incorrectly typed derived values.
• Updated the multi-lens research example with a run spec invariant that checks final citation IDs against mapped evidence source IDs.

v0.4.0

What's Changed

• Clarified the recommended first adoption path for Contract4Agents across the README, docs, tutorials, and examples.
• Improved first-run reliability for example evaluations by adding clearer fixture start points and trace output support.
• Updated example harnesses to record agent attribution for tool calls, improving evaluation trust and trace clarity.

v0.3.0

What's Changed

• Added static analysis for agent host_context dependencies, including validation that child agents can receive required context from parent agents or datasources.
• Added host_context details to compiled manifests and generated documentation.
• Added capability registry support via contract4agents.registry.json, with validation and drift checks against compiled contracts for tools, hosted tools, agents, output types, prompts, and host context.
• Added run spec evaluation support for host-owned multi-agent runs, including stage output validation and trace-based assertion checks.
• Improved datasource resolution behavior, including clearer handling for missing, ambiguous, cyclic, and denied datasource dependencies.

v0.2.0

What's Changed

• Added a richer OpenAI Agents SDK adapter with contract-based agent planning, factory helpers, output type registry generation, trace hooks, approval handling, and run-time assertion evaluation.
• Added support for OpenAI/provider-hosted tools, including openai.web_search, explicit hosted tool registries, and hosted-tool trace events.
• Added strict, versioned JSONL trace output with run IDs, event IDs, schema versions, known event types, and trace-loading diagnostics.
• Added guard plan generation for output conformance, denied tools, and approval-required tools, with adapter caveats for host-enforced or unsupported guard behavior.
• Added Pydantic interoperability for contract types, including optional CLI support for importing Python model types with --allow-python-imports.
• Added generated Pydantic output models for OpenAI structured outputs from compiled Contract4Agents schemas.
• Added host-callable assertion APIs for evaluating agent and run assertions outside the fixture runner.
• Added validation for malformed composition declarations, unknown composition targets, and composition entries without matching use agent dependencies.
• Added validation for unknown or incorrectly typed agent attributes, with hints for common mistakes like guard vs guards.
• Improved OpenAI tool name encoding so contract tool names with dots and underscores map safely and reversibly to SDK-safe names.
• Fixed nested native schema references so compiled schemas include needed $defs and nested output validation works correctly.
• Fixed trace monitor false positives by matching targets only against relevant trace fields and scoping monitor checks to the rule’s agent.
• Fixed datasource resolution to skip recursively unsatisfiable datasource candidates and choose valid dependency chains.
• Added safety checks to prevent compiler artifacts from being written into the project root or source-owned directories.
• Improved fixture verification so expected tool permissions are declared in fixture metadata instead of hard-coded.

v0.1.0

What's Changed

Initial public release of contract4agents: typed declarative contracts for AI agent systems.

This release includes the core contract DSL, parser, static checker, compiler, CLI, runtime trace helpers, eval and monitor runners, generated artifacts, OpenAI adapter, docs, and deterministic public examples for incident command, multi-lens research, and market research briefs.