Skip to content

CHANGELOG.md

Latest commit

 

History

History
366 lines (251 loc) · 12.3 KB

CHANGELOG.md

File metadata and controls

366 lines (251 loc) · 12.3 KB
Raw

Changelog

This changelog highlights user-facing changes in BabeL-O releases.

For full bilingual release notes, see docs/releases.

Unreleased

Added

  • Added bbl go --nexus-port-scan-attempts <n> to bound the local port scan used when the requested Nexus port is occupied (default: 12).

Changed

  • bbl go now prints a [bbl] ... notice when it switches to a different local port for the Nexus service.

Fixed

  • Fixed bbl go misdetecting any local service that answers /health with 2xx as a running Nexus (e.g. another dev server on the default port 3000), which caused Failed to allocate Go TUI session ... 404 API not found. The launcher now verifies Nexus identity (runtime: babel-o in the /health body) and auto-starts the managed Nexus on the first free neighbouring port when the requested one is occupied by a non-Nexus service.

v0.4.2 - 2026-07-24

Full release notes

Added

  • Added system keychain support for API key storage (macOS Keychain, Windows Credential Manager, Linux Secret Service). bbl config add stores keys in the keychain by default; use --plain for config file fallback.
  • Added bbl config init interactive wizard for first-time provider setup.
  • Added bbl config migrate to migrate plaintext API keys to the keychain.
  • Added bbl config audit to inspect credential storage locations.
  • Added ConfigManager.resolveSettingsAsync() to resolve settings with keychain-backed API keys (priority: env > keychain > config).
  • Added docs/guides/keychain-guide.md with credential security best practices.
  • Added docs/INSTALLATION.md with detailed installation instructions.
  • Added ProviderConfig.models field to support custom provider model definitions.

Changed

  • README now features a "Why BabeL-O?" section with 3 core differentiators, a "Quick Start (5 minutes)" guide, and simplified installation instructions.
  • All installation alternatives moved to the dedicated docs/INSTALLATION.md.
  • Go TUI watchdog grace increased from 60s to 120s (watchdog 300s) to prevent AskUserQuestion wait starvation during pre-question model latency.

Fixed

  • Fixed AskUserQuestion responses never reaching the model: the runtime now builds tool_result from finalOutput (user's answered selection) instead of result.output (raw pending_question payload).
  • Fixed Go TUI Esc-cancel in the AskUserQuestion overlay discarding the HTTP tea.Cmd, leaving waitForQuestionResponse to time out at 180s.
  • Fixed the question-response Zod schema rejecting Go nil slices (JSON null) for selectedIndices / selectedLabels, causing a 400 that prevented the response from being persisted to storage.

v0.4.1 - 2026-07-06

Full release notes

Added

  • Added runtime authorization and consent gates that separate tool need from user-authorized execution scope.
  • Added Go TUI, CLI one-shot, inspect-session, and trace visibility for authorization mismatches.
  • Added Everforest Light Soft theme coverage.

Changed

  • Improved soft-error and soft-timeout continuity so recoverable runtime states preserve clearer in-flight turn context.
  • Expanded Skill resource coverage and matcher quality for Agent Skills packages.

Fixed

  • Fixed bash tool startup when the user's login shell is incompatible by falling back to /bin/bash.
  • Reduced unwanted editing or over-confirmation when the user only gives a preference, status question, or meta-behavior question.

v0.4.0 - 2026-07-03

Full release notes

Added

  • Added Agent Skills package import/export interoperability while preserving BabeL-O's NormalizedSkill runtime shape.
  • Added Go TUI /skill list/show/validate workflow coverage and Skill overlays.
  • Added same-provider automatic retry for transient provider_unavailable and rate_limit failures, including retry lifecycle events.

Changed

  • Expanded Skill validation, resource coverage, conversion diagnostics, and matcher quality.
  • Added saved config and environment overrides for provider auto retry policy.
  • Moved more user-facing documentation into docs/guides/.

Fixed

  • Classified MiniMax transient unknown error, 999 (1000) and timeout 2066 failures as retryable provider availability errors.
  • Made provider retry waits stream provider_retry_scheduled before sleeping, so TUI/CLI countdowns appear immediately.
  • Cancelled provider retry waits now settle as REQUEST_CANCELLED without starting the next provider attempt.

v0.3.9 - 2026-06-23

Full release notes

Added

  • Added context-search pushdown and tokenized matching so long sessions can retrieve relevant historical events without overloading prompt assembly.
  • Added runtime and Nexus watchdog coverage for silent or aborted provider streams, including activeAgeMs diagnostics.
  • Added blocking layer-direction and coupling gates to CI.

Changed

  • Improved adaptive context-window selection and headroom preservation around recovery boundaries.
  • Split more Nexus, storage, runtime, and provider-loop responsibilities into smaller modules to reduce coupling.
  • Improved progressive streaming so provider text deltas are surfaced as they arrive instead of being buffered to the end of the turn.

Fixed

  • Preserved provider tool-call/tool-result message ordering across runtime replay and context assembly.
  • Propagated provider aborts through SSE readers so cancelled or timed-out streams settle cleanly.
  • Redacted context-observer payloads more aggressively and excluded internal hook, usage, and thinking events from default context-recent surfaces.

v0.3.8 - 2026-06-15

Full release notes

Changed

  • Moved Go TUI /model configuration progress messages out of the persistent transcript and into short-lived composer notices.
  • Generalized current-state verification guidance so provider, model, tool, config, session, workspace, test/build, MCP, and runtime checks stay evidence-backed.
  • Clarified WebSearch as an external public-information locator, not a source for private workspace facts or instructions.

Fixed

  • Displayed DeepSeek's declared 1M context window in Go TUI context chrome while still showing the effective runtime ceiling.
  • Kept soft-deny provider tool calls visible to the model while preserving the permission gate for write and execute actions.
  • Fed permission denials back to the provider as recoverable tool results so it can adjust instead of terminating the turn immediately.
  • Reused approved session-scoped provider rules to avoid repeated permission prompts for the same approved action.

v0.3.7 - 2026-06-14

Full release notes

Changed

  • Made bbl go the single production interactive TUI entrypoint.
  • Removed the legacy bbl chat TypeScript TUI from the release package.
  • Kept bbl run "<prompt>" for one-shot automation and scripting.
  • Updated CI and smoke checks to validate the Go TUI path.

Fixed

  • Preserved Go TUI permission response fields such as scope, rule, and feedback.
  • Applied trusted session approvals consistently within the active runtime session.

v0.3.6 - 2026-06-14

Full release notes

Fixed

  • Fixed portable bbl go launch on macOS after bbl go --check had passed.
  • Started bundled TUI binaries from the installed binary directory instead of a source checkout path.
  • Improved readiness checks and portable install regression coverage.

v0.3.5 - 2026-06-14

Full release notes

Changed

  • Switched the primary release artifact to a lightweight bbl-<platform>.tar.gz portable package.
  • Reused the user's system Node.js >= 22 instead of shipping a large Node SEA executable.
  • Kept standalone go-tui-* assets for debugging and compatibility.

Fixed

  • Restored the explicit API-key step in /model.
  • Treated pasted API keys as single-line secrets to avoid duplicated or corrupted input.
  • Routed mouse wheel events away from text fields.
  • Normalized composer height after returning from /model and /context.

v0.3.4 - 2026-06-13

Full release notes

Added

  • Added the /memory panel for MemoryOS status and memory candidate review.
  • Added stronger context, recovery, and working-set diagnostics.

Changed

  • Replaced hard-coded natural-language intent guidance with structured problemTarget and turn-policy metadata.
  • Improved timeout recovery, compact/restore validation, and provider replay.

Fixed

  • Fixed context, model, session, tool, and memory overlays so they render above the input area.
  • Hardened Go TUI layout restoration after panel exits.
  • Improved installer self-checks and release asset validation.

v0.3.3 - 2026-06-13

Full release notes

Added

  • Promoted the Go TUI as the documented production interactive client.
  • Added panel-driven /session workflows for creating, selecting, switching, and copying session IDs.
  • Added a current product screenshot and product-facing README refresh.

Changed

  • Refined the header, welcome panel, footer, input box, slash palette, context panel, tool overlays, and runtime activity line.
  • Made session-to-session messages typed collaboration context rather than hidden direct instructions.

Fixed

  • Improved model setup guardrails for API-key providers.
  • Reduced accidental input pollution from mouse selection and wheel behavior.
  • Reworked cancellation and recoverable conflict paths to ask the user instead of hard-cancelling ambiguous work.

v0.3.2 - 2026-06-09

Full release notes

Added

  • Added first-class SessionChannel collaboration with inbox, typed side-channel messages, channel graph diagnostics, activity overlay, and a Collaborate hub.
  • Added CLI commands for session inbox inspection and acknowledgement.

Fixed

  • Made interactive chat fail clearly under non-TTY conditions.
  • Kept standalone binary chat sessions alive while idle.
  • Tightened path-drift checks for ListDir, Grep, and Read.

v0.3.1 - 2026-06-08

Full release notes

Fixed

  • Fixed standalone binary startup failures caused by duplicate createRequire declarations.
  • Retained hardened installer behavior so incomplete downloads and HTTP error bodies cannot be installed as bbl.

v0.3.0 - 2026-06-07

Full release notes

Added

  • Added BabeL-O and KezhongKe logo assets to the README.
  • Added file attachment references, image metadata handling, symbol mentions, diagnostic mentions, and opt-in vim input mode.
  • Added optional bundled ripgrep support through @vscode/ripgrep.

Changed

  • Reorganized docs/nexus into active, reference, and archive layers.
  • Hardened installer downloads with retries, temporary files, size checks, and executable validation.

v0.2.9 - 2026-06-06

Full release notes

Added

  • Added persistent, governable Agent jobs with lifecycle metadata.
  • Added richer runtime metrics for provider invocations, AgentLoop steps, and AgentJob aggregation.
  • Added benchmark history, retry policy benchmarks, and runner comparison diagnostics.

v0.2.8 - 2026-06-03

Full release notes

Added

  • Added cache-aware context compaction and long-context utilization diagnostics.
  • Added a structured runtime diagnostics envelope.
  • Added quality gates including format:check, lint, coverage, build:smoke, deps:audit, and test:performance.

Changed

  • Improved /context, /status, slash/tool palettes, multiline input, paste state, and agent status rendering.

v0.2.7 - 2026-06-02

Full release notes

Added

  • Added MiniMax-M3 to the model registry.
  • Aligned model list syncing for autocomplete and interactive configuration.

Changed

  • Updated documentation language around BabeL-O as a generalized agent system.

v0.2.6

Full release notes

Added

  • Introduced Node.js SEA single-executable distribution.
  • Added a decoupled Nexus backend runtime architecture.
  • Added multi-agent Git worktree coordination, sandbox controls, path safety, HMAC command validation, and SQLite audit persistence.