0.134.1

• Support optional clientName config field for MCP servers to override the OAuth Dynamic Client Registration client_name (useful for servers that allowlist clients by name, e.g. Figma).

0.134.0

• Support including AGENTS.md files from parent directories of each workspace folder via new includeParentAgentsFiles config flag (disabled by default), ordered outermost parent first.
• Replace custom stderr-print logger with Logback/SLF4J: timestamps, log levels, chat-id MDC context, third-party noise suppression (root at WARN, eca at INFO), and proper cross-thread context propagation in future*. #253
• Fix MCP OAuth auto-discovery for servers that only return a 401 + www-authenticate challenge when probed with a valid JSON-RPC initialize request (e.g. Figma).
• Mark MCP servers as failed when the initialize handshake returns no result, instead of silently appearing as running.

0.133.6

• Bugfix: network.caCertFile (and clientCert/clientKey/clientKeyPassphrase) set via config.json were silently ignored due to a key-case mismatch between config normalization and the network reader; only the env-var fallbacks worked. #457

0.133.5

• Improve CPU usage while streaming tool-call arguments by reusing the prompt's tool list.
• Improve connection error messages from LLM providers. #457

0.133.4

• Bugfix: stop the infinite "Cannot run program 'kill'" liveness-probe log loop for sandboxed environments.

0.133.3

• Add unit and integration tests covering parent↔subagent end-to-end communication so regressions like the v0.133.1 spawn-agent breakage are caught automatically.
• Improve editor_diagnostics tool summary to show the target filename (e.g. Checking diagnostics: foo.clj) or Checking all diagnostics when no path is provided.
• Bugfix: preserve the chat's selected variant when changing model or agent on an existing chat (regression from per-chat scoping in v0.133.1).
• Native ECA tools now auto-resolve bare tool names like write_file to their canonical eca__... form, avoiding repeated failed retries when an LLM omits the native server prefix.
• Support Claude console subscription auth: dispatch x-api-key or Authorization: Bearer header based on configurable auth type.

0.133.2

• Fix subagent spawning being rejected by reserved server-managed chat ids.
• Fix OpenAI OAuth requests dropping built-in web search and image generation tools, and avoid replaying web search history artifacts on follow-up turns.

0.133.1

• Support client-generated chat ids: clients may now create the chatId themselves and send it on the first chat/prompt. eca-emacs#231.
• Per-chat scoping of model/agent changes: chat/selectedModelChanged and chat/selectedAgentChanged now accept an optional chatId; when provided the server scopes the persisted state and the resulting config/updated broadcast to that chat. The remote REST endpoints /chat/:chatId/select-model|select-agent|select-variant now forward the URL chat-id to those handlers (previously dropped). validate-client-chat-id is now public and rejects whitespace / control chars. eca-emacs#231.

0.133.0

• Add chat/promptSteerRemove notification for discarding a pending steer message before it is consumed at the next LLM turn boundary. Idempotent: silent no-op when no steer is pending.

0.132.1

• Support ask_user agent questions for remote REST/SSE clients (e.g. eca-web): the remote server now declares the askQuestion capability, broadcasts chat:ask-question over SSE with a generated requestId, and accepts answers via the new POST /api/v1/answer endpoint. Falls back to the JSON-RPC inner messenger when no SSE clients are connected, preserving editor behavior.