v2026.10.10-2

@code-yeongyu/senpi

Breaking Changes

Added

• /mcp with no arguments opens an interactive server manager in the TUI: compact server rows with state-specific actions (enable/disable, exposure, test, reconnect, sign in/out), plus tool, detail and log views. Only trusted global and project servers can be edited, and edits change just enabled/exposure in the existing config file. Untrusted project, imported, extension and skill servers are read-only (Details and Logs). Non-TUI clients still get the status text (#2716, #2747 by @wuhaoyujerry).

Changed

Fixed

• Goal continuations queued during compaction now resume after a stale or failed summary. Cancellation and context-limit checks still apply (#2778).
• Wake notices distinguish timer backstops from finished background work and label cache figures as cumulative estimates. Compaction logs identify the session (#2778).
• Bare model-family fallback no longer silently selects OpenGateway or Vercel AI Gateway when those providers have credentials. Like OpenRouter, these gateways require an explicit provider-qualified fallback selector (#2774).
• A project whose .senpi/codemode.json sets a setting that names an executable (today languages.pyInterpreter) now asks for project trust before the session starts, like a project mcp.json; before, such a project counted as trusted without asking, so the interpreter it named ran at session start (#2772).
• RPC model lists now reflect credentials added or removed by another session without reopening the session, fixing an empty desktop model picker after connecting a provider (#2769).
• The /btw side panel renders its answer as Markdown, the same way assistant messages render in the transcript. Headings, bold, inline code, lists, and code blocks used to show as raw ##, **, and backticks (#2766 by @nahwan-kim).

Removed

@code-yeongyu/senpi-ai

Breaking Changes

Added

Changed

Fixed

Removed

@code-yeongyu/senpi-agent-core

Breaking Changes

Added

Changed

Fixed

Removed

@code-yeongyu/senpi-codemode

Breaking Changes

Added

• @code-yeongyu/senpi-codemode/executable-settings.json lists the settings that name an executable run at session start (today languages.pyInterpreter). senpi's project-trust check reads it, so a project codemode file that sets one asks for trust, and a test fails if a new free-form string setting is neither on the list nor marked as not naming an executable (#2772).

Changed

Fixed

• Settings that were accepted but did nothing now take effect (#2763): kernelTools.enabled: false makes JavaScript tool(fn) and Python @tool refuse with tools_unavailable; languages.pyInterpreter makes the Python kernel run exactly that executable (a path that does not answer makes Python unavailable, with a warning naming the setting; one named by a project's own settings file is honored only in a trusted project); prompt.advertiseHelpers: true adds one line pointing at tool_schema('eval:helpers') to the eval description. Settings-file warnings (an unknown key, a fallback to defaults) now reach the user as a notice, or on stderr without a UI.

Removed

@code-yeongyu/senpi-tui

Breaking Changes

Added

Changed

Fixed

Removed