Skip to content

CHANGELOG.md

Latest commit

 

History

History
141 lines (130 loc) · 24.6 KB

CHANGELOG.md

File metadata and controls

141 lines (130 loc) · 24.6 KB
Raw

Changelog

All notable changes to Abigail are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Added

  • 2026-03-07 22:35 EST: Harden alpha release signing flow - normalize Tauri updater minisign keys to the Tauri 2 base64 format, accept both raw and base64 updater secret inputs, and allow fast alpha pre-releases to ship signed updater metadata before Windows Authenticode secrets are available
  • 2026-03-07 fix(crypto+topics): email-induced damage fully excised + protected-topic secrets for mentor-independent self-sufficiency (paper 22-27).
  • 2026-03-06 Skills strategy: Browser skill Playwright upgrade with persistent auth fallback for webmail and other OAuth-heavy flows.
  • 2026-03-05 [VAULT] Sentinel + isolated KEK + model-switch safety (eliminates AES-GCM wrong-key error forever)
  • 2026-03-05 23:50 EST: Add dedicated soul-forge worker module entrypoint and enforce forge mentor-approval gate before artifact persistence, with prompt/template updates reflecting approval requirement
  • 2026-03-05 23:35 EST: Add abigail-router::monitors::mentor_chat compatibility module and explicit selected-model subscriber registration API, wiring entity-daemon chat routes to register model subscriber identity on chat-topic
  • 2026-03-05 23:20 EST: Add phase-2 persistent topology wiring helper for registry path strings, use it in entity-daemon startup and watcher reprovision flow, and register builtin.devops_forge instruction mapping
  • 2026-03-05 23:05 EST: Update canonical skill topology and forge architecture spec to v1.2 with explicit out-of-band monitor flow, and sync planning/invariants/design references
  • 2026-03-05 22:35 EST: Add DevOps Forge worker with sandboxed skill authoring pipeline (topic.skill.forge request/response), superego gating, registry hot-reload writes, daemon startup wiring, and architecture/prompt documentation updates
  • 2026-03-05 21:55 EST: Add mentor chat monitor and out-of-band id/superego/memory layer with chat-topic preprompt enrichment, selected-model subscriber wiring, and startup monitor registration
  • 2026-03-05 21:35 EST: Add persistent startup skill topology provisioning from registry.toml with request/response topics, per-skill subscribers, registry hot-reload reprovisioning, and forge registry entry
  • 2026-03-05 21:20 EST: Formalize skill topology and Forge architecture docs, add canonical ARCHITECTURE_SKILL_TOPOLOGY_AND_FORGE spec, update planning/invariants artifacts, and add architecture flow diagram
  • 2026-03-05 20:30 EST: Add regression test for provider-config-changed selector refresh and update README/CLAUDE/AGENTS plan to mark Phase 0 complete and Phase 4b mentor monitor as next
  • 2026-03-04 20:00 EST: Provider key propagation + model registry sync — emit provider-config-changed event on key store, synchronous model discovery, auto-activate first LLM provider, frontend event listener for live dropdown refresh, fix ChatRendering test multiple-match assertion
  • 2026-03-04 19:30 EST: Remove legacy routing config panel and fix provider/model dropdowns — delete ~520 lines of unused ConfigStep UI (provider setup buttons, routing mode selectors, CLI server controls), fix CliOrchestrator dropdown to show static provider label instead of irrelevant model selectors, filter CLI providers from model registry, suppress model_override in CLI mode
  • 2026-03-04 18:45 EST: Fix provider/model dropdown pipeline — thread model_override from UI through gateway/Tauri/entity-chat to router, wire provider dropdown to use_stored_provider backend, remove dead ForceOverride code, filter non-LLM providers from dropdown
  • 2026-03-04 10:30 EST: Fix crypto bootstrapping crash on fresh install — return signing key directly from generate_master_key to avoid KEK mismatch when OS keyring store fails, add vault format detection in load_master_key to prevent AES-GCM data being fed to DPAPI
  • 2026-03-04 08:40 EST: Fix Windows installer startup failure — bundle WebView2 bootstrapper in NSIS installer, replace silent startup panics with native OS error dialogs (MessageBoxW on Windows) so users see what went wrong instead of the app silently vanishing
  • 2026-03-04 00:05 EST: Wire provider/model dropdown to chat — add get_force_override/set_force_override Tauri commands, force_override state, pass model_override through stream_chat_pipeline to route_unified
  • 2026-03-04 07:30 EST: Theme engine — three built-in themes (Modern Clean, Phosphor Terminal, Classic Desktop) with hive-level default + per-entity override, ThemeDrawer on SoulRegistry, Appearance tab in Sanctum, birth flow theme selection, ~250 hardcoded color migrations to semantic tokens, expanded CSS token system (typography, border-radius, semantic colors, overlays)
  • 2026-03-04 05:05 EST: Integrated skills upgrade — backend confirmation enforcement (SkillExecutor), shared secret redaction (redact_secrets in core), strict tool argument validation, normalized tool result semantics (shell/git success=false on failure), manifest-driven daemon secret sync, native skill parity in entity-daemon (all 15 skills), dynamic skill method-aware confirmation defaults, Iggy connection configurable in AppConfig with DaemonManager passthrough, standardized 5-topic bootstrap, and fixed stale integration instructions
  • 2026-03-03 20:55 EST: Cross-platform vault overhaul — replace DPAPI-only encryption with AES-256-GCM envelope + HKDF-scoped key derivation, hybrid unlock (OS keyring / passphrase fallback), rewire SecretsVault/Keyring/encrypted_storage to new crypto core, add shared RESERVED_PROVIDER_KEYS, harden daemon secret namespace validation
  • 2026-03-03 17:50 EST: Fix ChatInterface null storedProviders in tests — guard against null/undefined from get_stored_providers mock
  • 2026-03-03 16:33 EST: Phase 4 Autonomous Job Delegation — fix provider/model dropdowns (whitelist chat models, merge vault providers), chat page job badge with live event stream, ego delegation tools (submit_background_job/get_job_result/list_my_jobs), capability-aware routing (ImageGeneration/AudioGeneration/VideoGeneration/Transcription), ExecutionMode Direct for skip-LLM skill dispatch, cron self-management tools, OrchestrationPanel event-driven updates, V6 schema migration
  • 2026-03-03 17:30 EST: Complete Phase 3 Simplification, Observability, and UX — 3a: remove Council/TierBased routing (~2000 lines); 3b: Jobs tab with live JobQueue list and recurring templates; 3c: system prompt viewer in Identity/Soul tab; 3d: replace ForgePanel with Nerve Center topic monitor; 3e: Data Explorer with memory browse/search/sessions/stats tabs
  • 2026-03-03 14:30 EST: Complete Phase 2c Memory and Integration — daemon-test-harness crate (HiveDaemonHandle, EntityDaemonHandle, TestCluster), hive-daemon and cross-daemon E2E tests, daemon-client crate (HiveDaemonClient, EntityClient with SSE), RuntimeMode config (InProcess/Daemon), DaemonManager for managed daemon lifecycle, chat_stream/list_skills/store_secret dual-mode delegation, DiagnosticsPanel runtime mode toggle
  • 2026-03-03 13:15 EST: Complete Phase 2b Skill Creation — Dynamic API authoring docs, recursive discover (nested subdirs), SkillsWatcher for skill.toml+JSON hot-reload with register/unregister, StreamBroker skill events, Tauri watcher wiring with UI toast notifications
  • 2026-03-04 16:30 EST: Add file upload and drag-and-drop to ChatInterface — paperclip file picker, drag overlay, attachment preview strip, save-to-entity-docs persistence, and get_entity_documents_path/save_to_entity_docs Tauri commands
  • 2026-03-03 23:30 EST: Add BackupManagementSkill — entity self-restore: list/preview/import conversation turns and memories from backup SQLite databases into running MemoryStore
  • 2026-03-04 14:00 EST: Fix 6 runtime bugs — author_skill secrets declaration, email SMTP config, filesystem allowed_roots expansion, OpenAI non-chat model filtering, perplexity reserved key, and provider-to-skill secret sync
  • 2026-03-03 19:10 EST: Add OllamaDrawer (left) and ProviderDrawer (right) slide-out panels to SoulRegistry for pre-birth Ollama management and provider/CLI configuration
  • 2026-03-03 17:55 EST: Add inline Provider + Model dropdowns to ChatInterface toolbar for instant model switching without leaving the conversation
  • 2026-03-03 15:30 EST: Address 6 residual risks — wire server-side cancel for SSE streams (daemon + Tauri), expose governance constraints via routes and Tauri commands, remove legacy blocking chat command, deprecate target field across full stack, expose job scheduler via Tauri commands, add entity-cli skill-sign subcommand for signed allowlist entry generation
  • 2026-03-03 13:40 EST: Fix Ollama startup lifecycle — resolve premature model_loading state, progress bar visibility for non-typewriter paths, double identity-check, pull button port mismatch, CI curl hardening, and fix download URLs to extract from archives (zip/tgz) instead of non-existent standalone binaries
  • 2026-03-02 12:15 EST: Fix "Unsupported 16-Bit Application" error — validate bundled Ollama is a 64-bit PE before spawning, fall back to system Ollama when invalid
  • 2026-03-02 23:20 EST: Fix startup flow — bundle Ollama in release-fast workflow, fix pull button parameter mismatch, show loading screen instead of skipping when Ollama unavailable
  • 2026-03-02 21:40 EST: Fix release workflow version auto-increment failing on suffixed tags (e.g. v0.0.65-fast.4) — filter to clean semver tags only
  • 2026-03-02 20:30 EST: Fix Ollama auto-connect skipping model download — add list_ollama_models backend command, guard auto-connect on installed models, replace button list with dropdown model picker (installed/recommended/custom)
  • 2026-03-02 16:00 EST: Topic-based decomposition — kill EventBus for StreamBroker, async memory consumer, cron scheduling in JobQueue, lean orchestrator prompt, instruction topic-affinity, unified router method, async tool-use delegation with depends_on job chains, AgenticEngine StreamBroker events, conscience monitor topic consumer, council job fan-out
  • 2026-03-01 01:00 EST: Fix model loading screen progress (serde event format mismatch) and auto-skip Ignition stage when managed Ollama is running with model ready
  • 2026-02-28 23:30 EST: Bundle Ollama with ABNORMAL BRAIN loading screen — auto-start managed Ollama, pull Llama 3.2 3B on first launch with typewriter-flicker loading animation, graceful shutdown on exit, release workflow downloads and codesigns Ollama per-platform
  • 2026-02-28 21:55 EST: Fix macOS notarization — codesign bundled abigail-keygen binary with hardened runtime and secure timestamp before Tauri build
  • 2026-02-28 16:00 EST: Add sub-agent job queue Phase 1 — abigail-streaming crate (StreamBroker trait + MemoryBroker), abigail-queue crate (JobQueue with SQLite + dual-layer event publishing), V3 migration for job_queue table
  • 2026-02-28 14:30 EST: Add Apple notarization to macOS release workflow and update CLAUDE.md with 5 missing crates, soul crystallization, agent backup/restore, and 12 frontend components
  • 2026-02-28 12:30 EST: Fix birth cycle CLI indicators all turning green when one is enabled — only show active for explicitly activated providers
  • 2026-02-28 04:55 EST: Add macOS Developer ID code signing to release workflow — import certificate into CI keychain, pass signing identity to Tauri build, clean up keychain after build
  • 2026-02-27 14:30 EST: Rename skill-proton-mail to generic skill-email — remove Proton-specific defaults, make imap_host/smtp_host required, use Network=Full permissions
  • 2026-02-28 02:00 EST: Agent backup/restore system, shared MemoryStore, chat turn persistence, confirmation modals, lenient tool param parsing, and Proton Mail skill auto-reinit on secret change
  • 2026-02-28 01:00 EST: Ignition page auto-detects all providers on mount, highlights available tabs with green indicators, auto-focuses sole available option, remove redundant CLI Tools footer
  • 2026-02-28 00:30 EST: Compress CLI system prompt (~15-41KB → ~1.5KB) with temp file spillover, budgeted instruction injection (max 1 / 2048 bytes for CLI), compact grouped tool list, and reduced max-turns to 5
  • 2026-02-27 23:15 EST: Update docs for CliOrchestrator auto-detection, remove stale --allowedTools refs, mark completed Phase 2a/2c roadmap items, fix CliPermissionMode doc comments
  • 2026-02-27 22:30 EST: Fix intermittent CLI spawn OS error 206 on Windows — replace broken --allowedTools args (entity tool names, not CLI tool names) with --dangerously-skip-permissions, auto-detect CliOrchestrator routing mode for CLI providers to skip meaningless tier scoring/complexity classification
  • 2026-02-27 19:15 EST: CLI session management, memory auto-archive, and encrypted portable archives — pipe Claude CLI prompt via stdin to fix Windows 32K limit, add --resume/session_id for multi-turn, conversation_turns table and auto-archive in entity-daemon/Tauri, build_memory_context and prompt memory awareness, X25519+AES-256-GCM .abigail export to Documents/Abigail/archives/, restore via recovery key, auto-export every 50 turns
  • 2026-02-27 21:00 EST: Birth/Entity chat pipeline separation — extract BirthChatEngine to abigail-birth crate with scripted no-LLM fallback, move key detection to abigail-core, add birth conversation persistence, slim Tauri commands to thin wrappers, add best_available_provider() to router
  • 2026-02-27 15:30 EST: Full agentic CLI integration — CLI auth verification (official binary + auth status), CliOrchestrator routing mode, CLI Quick-Start in Ignition stage, rich command flags (--append-system-prompt, --output-format, streaming), CLI-optimized system prompt builder, routing mode selector in ChatInterface
  • 2026-02-27 04:10 EST: Apply cargo fmt for CI lint gate
  • 2026-02-27 03:50 EST: Enrich CLI REST server chat pipeline — system prompt, tool wiring, tool-use loop, full metadata; fix status to read live router state; register all 15 native Rust skills at Tauri startup
  • 2026-02-26 22:20 EST: Add Insights troubleshooting panel — CLI REST server toggle with connection info, in-memory log ring buffer with reloadable tracing filter, live log viewer with stream/pause/clear/export, save-to-file via dialog
  • 2026-02-26 20:15 EST: Routing simplification — make ExecutionTrace authoritative source of truth, add SelectionReason enum, implement explicit Council execution path, staged force overrides in Forge, diagnose_routing endpoint/command, operator runbook and migration docs
  • 2026-02-26 16:30 EST: Entity-First Attribution — normalize chat-facing labels so Id never appears as conversational actor, map id(...) to "local" in UI and provider_label(), add entity-identity self-report rule to runtime prompt
  • 2026-02-26 16:00 EST: Execution Truth Refactor — add execution_trace DTO and traced router paths, thread trace through entity-chat and Tauri/daemon responses, show configured+executed attribution and fallback chain in chat UI
  • 2026-02-26 19:45 EST: Add entity runtime self-awareness — inject RuntimeContext (provider, model, tier, entity name) into system prompt, filter phantom skill instructions against registered skills, track provider change timestamps in config (schema v19)
  • 2026-02-26 16:15 EST: Fix Tauri skill/secrets flow — register ProtonMailSkill at startup, unify secret namespace validation, bootstrap skill instructions into data_dir, add live E2E probe and PowerShell runner
  • 2026-02-26 02:30 EST: Fix OpenAI/xAI/Google/local tool-name sanitization — extract shared sanitize_tool_name and build_tool_name_map to provider.rs, apply to all OpenAI-compatible providers (build_tools, build_messages, and reverse-map in complete/stream), consolidate Anthropic's local copy
  • 2026-02-26 01:30 EST: Align streaming router path — refactor route_stream() to use target_for_mode() matching all other routing methods, remove dead EgoPrimary fallback code, extract shared stream_chat_pipeline() and provider_label() into entity-chat eliminating ~90 lines of duplication between Tauri and daemon
  • 2026-02-25 22:15 EST: Add graceful local LLM error handling — parse JSON error bodies for friendly messages, add Id→Ego fallback on all router paths, and surface actionable guidance in chat UI for "no model loaded" and connection errors
  • 2026-02-25 17:40 EST: Implement streaming chat (Tauri events + daemon SSE), wire SkillFactory runtime auto-registration with DynamicApiSkill output, add E2E parity tests with mock providers across entity-chat/daemon/frontend
  • 2026-02-24 23:00 EST: Implement SMTP client, wire ProtonMailSkill send_email, add live email E2E tests (entity-chat), fix Anthropic tool-name sanitization for all skill tool-use, extract HiveOperations trait for Tauri/daemon parity
  • 2026-02-24 17:30 EST: Add tabula rasa UAT harness with STARTTLS IMAP support, entity-daemon --data-dir isolation, Hive-to-entity secret sync, and automated 9-stage test pipeline
  • 2026-02-26 12:00 EST: Remove Superego from entity entirely; add ChatMemoryHook at entity memory persist as sole hook for future Hive/Superego use
  • 2026-02-26 01:00 EST: Implement tier-based model routing overhaul — 3-tier model selection (Fast/Standard/Pro) with complexity scoring, force overrides (pin tier/model/provider+tier), dynamic model registry with provider API discovery and 24h TTL caching, per-request model_override in CompletionRequest, remove IdPrimary routing mode, add tier metadata to ChatResponse, and build full Forge UI with tier assignment grid, force override controls, and threshold tuning
  • 2026-02-25 06:00 EST: Extract shared chat engine into entity-chat library crate, eliminating ~200 lines of duplicated pipeline code between entity-daemon and Tauri app so CLI and GUI use a single engine
  • 2026-02-25 04:30 EST: Fix HiveManagementSkill tool schemas causing OpenAI 400 errors, add defensive schema validation in build_tool_definitions, implement dynamic model discovery (discover_models API + Hive endpoint + startup diagnostics)
  • 2026-02-25 03:00 EST: Strip chat flow to standard minimum — remove redundant tool awareness markdown, dead "Blocked:" postfix, and write-only memory persistence from both GUI and CLI pipelines (10→6 steps GUI, 8→5 steps CLI)
  • 2026-02-25 02:00 EST: Unify GUI and CLI chat flows — replace streaming chat_stream with non-streaming chat command mirroring entity-daemon pipeline (tool-use loop, tool awareness, memory persistence), delete ~650 lines of superfluous code, simplify ChatInterface.tsx frontend
  • 2026-02-24 15:30 EST: Wire abigail-memory into entity-daemon with REST endpoints, chat persistence, memory CLI commands, tool-use loop, skill auto-loading, and skill scaffolding CLI
  • 2026-02-23 22:00 EST: Add full chat pipeline to entity-daemon (system prompt, message sanitization, tool awareness, deduplication, risk clarification) for parity with Tauri app
  • 2026-02-23 18:30 EST: Disable Tauri updater artifact signing (createUpdaterArtifacts: false) to fix release build failure from malformed signing key
  • 2026-02-23 18:00 EST: Update all documentation to reflect Hive/Entity daemon architecture; remove superfluous MVP_SCOPE, PHASE1_AGILE_PLAN, and prompt-routing-review docs
  • 2026-02-23 16:00 EST: Implement Hive/Entity separation Phase 1 — seven new crates (hive-core, entity-core, abigail-identity, hive-daemon, entity-daemon, hive-cli, entity-cli) splitting the Tauri monolith into separate control-plane and agent-runtime HTTP daemons
  • 2026-02-22 19:00 EST: Create abigail-hive crate as single authority for secret resolution and provider construction, moving builder functions out of router and Tauri app
  • 2026-02-22 01:30 EST: Harden release signing key pipeline to auto-strip whitespace from base64, validate encoding, and pass sanitized key to build step
  • 2026-02-22 00:10 EST: Fix CI lint gate by correcting import ordering in config.rs
  • 2026-02-21 23:20 EST: Fix native birth/chat provider reliability and improve Forge feedback/readability while hardening browser harness tests to catch real-world provider failure and recovery behavior.
  • 2026-02-21 19:30 EST: Fix TAURI signing-key validator escaped-newline detection so CI preflight tests pass for valid minisign secret formats.
  • 2026-02-21 19:15 EST: Add lock-safe MCP server URL resolution and focused clippy guard to catch lock-across-await regressions in abigail-app CI.
  • 2026-02-21 15:40 EST: Add release signing-key preflight validation and CI regression test to catch malformed TAURI updater secret format before Tauri bundling.
  • 2026-02-21 15:00 EST: LLM routing and skills verification — unified router rebuild/superego, trust gating and capability envelope in execution, qualified tool resolution, routing tests, and vision chunk alignment (birth, identity, chat, forge, skills).
  • 2026-02-21 12:00 EST: Refresh documentation baseline by aligning README, CLAUDE guidance, release process, GitHub settings checklist, and environment notes with current CI/release workflows.
  • 2026-02-20 17:30 EST: Massive cleanup and rebranding to Sovereign Entity model: modularized Tauri backend, pruned legacy routing/stubs, rebranded UI to Soul Registry and Sanctum, and implemented agentic Recall memory tool.
  • 2026-02-19 19:14 EST: Added 10 new skill crates (knowledge-base, git, code-analysis, database, calendar, document, notification, image, clipboard, system-monitor) + browser instruction-only skill, permission parser fix, productivity subagent, and expanded file_ops/privacy subagent capabilities
  • 2026-02-19 20:45 EST: Implemented refined lightweight router – Fast Path (Id+Ego+Context) + out-of-band Superego/Trust conscience to eliminate lock-ups
  • 2026-02-19 13:34 EST: Added abigail-auth crate (Phase 1) — AuthProvider trait, StaticToken/BasicAuth providers, in-memory TokenCache, AuthManager registry with 11 tests
  • 2026-02-19 19:02 EST: Add preloaded integration skills (GitHub, Slack, Jira) with auth framework wiring — embedded DynamicApiSkill configs, AuthManager in AppState, versioned bootstrap, get_integration_status/store_integration_credential commands, check_integration_status LLM tool, instruction registry entries
  • 2026-02-19 23:30 EST: Fix credential storage refusal (conscience allowlist + enriched safety prompt + configure_email LLM tool) and add abigail-cli crate with CLI subcommands + REST troubleshooting API
  • 2026-02-20 00:15 EST: Register ProtonMailSkill in runtime, add get_system_diagnostics LLM tool, troubleshooting instruction keywords, and fix skill-proton-mail compile errors
  • 2026-02-19 14:30 EST: Unified UI theme system across all 20 components — chat bubbles, softer backgrounds, focus rings, scrollbar theming, ARIA accessibility, and zero hardcoded gray/blue colors
  • 2026-02-20 10:30 EST: UI overhaul — replace single-line input with auto-growing textarea, remove top tab bar, add slide-out Forge drawer with 10 sub-tabs, expose get_system_diagnostics as Tauri command, add DiagnosticsPanel
  • 2026-02-20 14:30 EST: Add Tauri auto-update plugin with UpdateNotification banner, fix NSIS backup for Hive multi-agent files, add SQLite migration framework with schema_versions table, and UPGRADE.md documentation
  • 2026-02-20 15:30 EST: Add CLI provider adapter for Claude Code, Gemini CLI, OpenAI Codex CLI, and xAI Grok CLI as Ego routing backends
  • 2026-02-20 16:00 EST: Add CLI provider options to birth Ignition/Connectivity stages, post-birth config menu, and LLM tool schema
  • 2026-02-20 19:30 EST: Fix CI gate by committing missing auto-updater deps, schema migration code, capabilities, and UpdateNotification component
  • 2026-02-19 10:55 EST: Public release readiness: LICENSE, CONTRIBUTING, CODE_OF_CONDUCT, SECURITY, CI workflows, CodeQL, and CODEOWNERS

Changed

  • 2026-02-18 23:04 EST: Enhanced README.md with badges, system requirements, and troubleshooting
  • 2026-02-18 01:36 EST: Updated .gitignore with additional patterns for generated and data files

0.0.1 - 2026-02-03

Added

  • Initial release of Abigail desktop agent
  • Interactive birth flow with staged onboarding
  • First-run Ed25519 signing key generation with one-time private key presentation
  • Constitutional document signing and verification (soul.md, ethics.md, instincts.md)
  • Local LLM discovery and manual connect for Ollama/LM Studio-compatible endpoints
  • In-app API key vaulting and validation for cloud/model/search providers
  • Dual persona UI modes (surface chat and Forge mode toggle)
  • Id/Ego routing: local LLM (Id) for routine queries, cloud LLM (Ego) for complex queries
  • Skill-based tool execution with web-search capability
  • DPAPI-encrypted secrets storage on Windows
  • Cross-platform builds: Windows (NSIS), Ubuntu (deb), macOS (dmg universal binary)
  • npm CLI installer (npx abigail-desktop)
  • Docker development and build containers
  • Security audit CI (cargo audit, npm audit)
  • Dependabot configuration for Cargo, npm, and GitHub Actions