v0.9.0

What's Changed

• chore(api): enforce public any-field contract and add coverage by @Viking602 in #15
• fix: remediate runtime stability risks by @Viking602 in #16
• feat: live agent streaming + reference multi-agent schedulers by @Viking602 in #17
• feat(multiagent): add typed DAG orchestration graph with bounded concurrency by @Viking602 in #18
• feat(multiagent): field-level fan-in, deep schema checks, team-level streaming by @Viking602 in #19
• feat(agent,worker): bound the agent loop with per-task budgets and typed failure classification by @Viking602 in #20
• fix(provider/anthropic): round-trip structured content blocks and thinking signatures by @Viking602 in #21
• fix(agent): harden loop failure paths (panic isolation, partial-trace, guardrail classification) by @Viking602 in #22
• docs(adr): translate ADR-001..008 to English by @Viking602 in #23
• feat(agent): self-sufficient agent layer — neutral Spec, model resolver, subagent-as-tool by @Viking602 in #24
• feat(workflow): add workflow modeling layer; rename scheduler driver to cron by @Viking602 in #25
• docs(readme): rewrite as a why-first overview by @Viking602 in #26
• feat(eval): redesign evaluation framework around EvalCase/Harness by @Viking602 in #27
• feat(coding): sandboxed coding agent with hashline edit protocol by @Viking602 in #28
• feat(agent): wire StepPolicy and ContextManager.Compact into the loop by @Viking602 in #29
• refactor: land 2026-06-10 architecture review fixes by @Viking602 in #30
• feat!: close v0.8.0 storage debt and reliability gaps by @Viking602 in #31

Full Changelog: v0.8.1...v0.9.0