a2a-bridge v0.2.1

Fixed

• Containerized and already-logged-in Codex agents can declare pre_authenticated = true, preventing
  the bridge from re-invoking codex-acp's interactive ChatGPT browser login before session/new.
  Shipped Codex configs now use the setting, restoring gpt-5.6-sol model/effort selection and prompt
  turns with codex-acp 1.1.2 in browserless containers.

a2a-bridge v0.2.0

Changed

• Relicensed Apache-2.0 → AGPL-3.0-only, while the project had a single copyright
  holder, plus a Contributor License Agreement (CLA) and a CONTRIBUTING.md stating the
  "maintained, not (yet) supported" stance (45bf05b).
• Wave 1 — runtime & CI hardening (0d4d12c): SQLite opened with WAL +
  synchronous=NORMAL + busy_timeout (warn-not-fail); the Configuring claim-state
  fix so lazy agent spawn no longer holds the registry lock across resolve; worktree
  git invocations moved off the async runtime onto tokio::process; CI toolchain
  pinned to 1.94.0 with the workspace MSRV inherited by every manifest.
• Wave 2 — identity & docs (db4a8b3): README rewritten to current capability
  (command table, crate table, troubleshooting, sample output); 172 one-shot
  dev-process artifacts moved (pure renames) into docs/history/, shrinking the
  workflow-artifact hygiene allowlist 208 → 37; ADR-0032 sandbox tier model plus
  example tier presets; AGENTS.md/onboarding.md synced to the current CLI.
• Wave 3 — CLI polish, doctor, A2A wire safety (18e1c5a): uniform --help across
  every subcommand via dispatcher-level interception; a2a-bridge doctor, a
  read-only bounded preflight (9 checks, host-vs-sandbox aware, --json); A2A golden
  wire fixtures pinning the a2a-lf SDK boundary (redaction, TaskSpecInvalid
  passthrough, ordered SSE frame contract).

Removed

• BREAKING: silent config auto-write removed. Bare a2a-bridge / serve / mcp
  invocations that can't resolve a config now error instead of scaffolding one —
  init is the only command that writes a config file.