v0.5.0 - Turn-as-Unit SSOT + 200-line discipline

Highlights

Architecture enforcement (8-phase plan)

• Move TurnTracker into loopal-context so TurnStore mutators are pub(crate) — runtime crates cannot bypass the tracker at compile time.
• Retire ContextStore::from_messages; tests seed history via loopal_test_support::seed_history.
• tests/architecture_boundary_test.rs greps for cross-layer wire-type leaks (MessageRole/ContentBlock outside provider/display/context layers).
• Wire-only mutations (microcompact, condense_server_blocks) now flow through TurnTracker::with_wire_mut so the ephemeral path also honors the SSOT contract.
• Re-derive open ToolBatch step on TurnTracker::new/replace_store so resume-mid-batch correctly routes update_tool_state.

File-size + comment discipline

• 10 source files split into directory modules: turn_event_store, turn_store, turn_tracker, turn_projection, request_turns, compaction, compact_rehydrate, ingestion, turn_degradation, resolver.
• Sibling extractions: SettleSignal, mcp_settle, session_start_prompt.
• All src/**/*.rs files ≤200 lines.
• Module-level //! blocks and /// docs that restate signatures stripped per HARD RULE — code is SSOT.

Dead-code removal

• Dropped provider-api::Middleware trait, ContextStore pipeline / config_refresh / degradation / ingestion middleware, governance::compensation.

Breaking changes

• loopal_runtime::agent_loop::turn_tracker moved to loopal_context.
• ContextStore::{from_messages, clear, truncate} removed.
• Internal middleware modules retired (see above).

Stats

49 commits since v0.4.0 · Net -2904 lines in #183 alone.

Full diff: v0.4.0...v0.5.0

What's Changed

• fix(goals): trigger continuation kickoff on Create/Resume/Extend control by @yishuiliunian in #138
• refactor(permission): decompose into orthogonal Permission × Decision axes by @yishuiliunian in #139
• refactor(decision): split Auto into Classifier and Agent with race-to-respond UX by @yishuiliunian in #140
• fix(events): repair /clear and seal Control→Event→View drift by @yishuiliunian in #141
• feat(memory): codify SNR + latent structure as Prime Axioms by @yishuiliunian in #142
• refactor(git): move gitignore SSOT into .loopal/.gitignore by @yishuiliunian in #143
• refactor(view-state): silence allow-path decision echoes by @yishuiliunian in #144
• fix(logging): never delete .log files held by live processes by @yishuiliunian in #145
• fix(tools): restore WebSearch and make Fetch actually download pages by @yishuiliunian in #146
• feat(prompt): introduce core/soul.md with five Prime Axioms by @yishuiliunian in #147
• feat(askuser): strict schema validation + multi-question segment UI by @yishuiliunian in #148
• fix(tui): guard render paths against tiny terminals by @yishuiliunian in #149
• fix(spawn): split permission_mode and decision_mode into independent params by @yishuiliunian in #150
• refactor(arch): 9 architecture debts + 5 review cycles + 200-line split by @yishuiliunian in #151
• refactor(scheduler): async save worker + test consolidation by @yishuiliunian in #152
• fix(scheduler): remove wait_idle from startup path by @yishuiliunian in #153
• feat(provider): add OPENAI_BASE_URL env var and GPT-5.5 model by @yishuiliunian in #154
• refactor(tools): split ReadPdf and ReadHtml out of Read tool by @yishuiliunian in #155
• fix(tools): treat empty process_id as absent in Bash tool by @yishuiliunian in #156
• refactor(tools): TypedTool trait + input/schema normalization by @yishuiliunian in #158
• fix(provider): preserve OpenAI reasoning item for web_search_call multi-turn by @yishuiliunian in #159
• fix(tool-api): ensure object schemas always include properties field by @yishuiliunian in #160
• fix(provider): flush each OpenAI reasoning item as independent Thinking block by @yishuiliunian in #161
• feat(tui): persist /model picker selection + fix release-event double-fire by @yishuiliunian in #162
• feat(vault): zero-trust secret subsystem so plaintext never reaches LLM by @yishuiliunian in #163
• refactor(cli): bring vault into clap subcommand tree + arg help by @yishuiliunian in #164
• refactor(goal): remove token budget subsystem by @yishuiliunian in #165
• feat(bg-task): per-session log files + actor-based process supervision by @yishuiliunian in #166
• feat(goal): allow Complete → Active reopen so model can recover from premature complete by @yishuiliunian in #167
• fix(ipc): typed RpcError + auto-respond contract so missing-session fails fast by @yishuiliunian in #168
• refactor(goal): drop barren-driven auto-complete by @yishuiliunian in #169
• feat(soul): add Axiom 6 — see systems, not parts by @yishuiliunian in #170
• feat(tui-panel): unify Cron/Task/BgTask detail sub-pages with stop action by @yishuiliunian in #171
• fix: scroll slash command menu + isolate cleanup_orphans flaky by @yishuiliunian in #172
• refactor(governance): split TurnObserver into Governance + TurnHook with 6 enforced invariants by @yishuiliunian in #173
• feat(compact): three-tier defense + boundary marker + API token feedback for 1M-context sessions by @yishuiliunian in #175
• feat(tool-image): add ReadImage tool + multimodal image channel for LLM by @yishuiliunian in #174
• feat(mcp): non-blocking startup + sub-agent proxy + actionable diagnostics by @yishuiliunian in #176
• fix(tui-askuser): Enter confirms default option without prior ↑↓ by @yishuiliunian in #177
• refactor(tools-fs): typed paths + batch IO + virtual-FS ApplyPatch + actionable errors by @yishuiliunian in #178
• feat(hub): move Vault + MCP to Hub process with sharing model by @yishuiliunian in #179
• feat(agent-loop): ContinuationGate + DegenerationDetector + request_idle by @yishuiliunian in #180
• refactor(bootstrap): root-cause hub-only handshake deadlock via typestate + IpcBudget by @yishuiliunian in #181
• fix(agent-loop): /goal replace + request_idle turn-end + intercept emit invariant by @yishuiliunian in #182
• refactor: enforce Turn-as-Unit SSOT + 200-line file discipline by @yishuiliunian in #183

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

• feat(prompts): proactively trigger explore sub-agent for codebase research by @yishuiliunian in #129
• fix(spawn): propagate --no-sandbox to all sub-agents by @yishuiliunian in #130
• feat(askuser): inline UI + structured response protocol by @yishuiliunian in #131
• feat(tools): scenario-aware truncation + fetch LLM refiner by @yishuiliunian in #132
• refactor(runtime): turn state machine + decouple prefill from thinking by @yishuiliunian in #134
• refactor(hub): standalone Hub subprocess + detach/kill lifecycle by @yishuiliunian in #133
• feat(goals): thread-level goal tracking with auto-continuation and budget accounting by @yishuiliunian in #135
• fix(goals): unblock /goal pipeline and unify command reply channel by @yishuiliunian in #136
• refactor(cli): split passthrough by consumer + reflective completeness tests by @yishuiliunian in #137

Full Changelog: v0.3.0...v0.4.0

v0.3.0

What's Changed

• feat(provider): add Claude Opus 4.7 and set as default model by @yishuiliunian in #118
• feat(tui): add Crons panel for scheduled cron jobs by @yishuiliunian in #119
• feat(scheduler): add durable persistence for cron tasks by @yishuiliunian in #120
• fix(status-bar): emit explicit Running event on turn start by @yishuiliunian in #121
• fix(status-bar): keep spinner/timer alive across turn boundaries by @yishuiliunian in #122
• feat(tui): panel section headers + fix focus arrow residue by @yishuiliunian in #123
• refactor(protocol): NAT-style cross-hub address routing by @yishuiliunian in #124
• refactor(agent): event-driven symmetric session-scoped state for cron/task by @yishuiliunian in #125
• refactor(spawn): split sub-agent spawn IPC into InHub vs CrossHub by @yishuiliunian in #126
• feat(inbox): event-driven visibility for inter-agent messages by @yishuiliunian in #127
• refactor(hub): event-driven multi-UI architecture by @yishuiliunian in #128

Full Changelog: v0.2.0...v0.3.0

v0.2.0

What's Changed

• refactor(sandbox): remove OS-level write restrictions, shift to permission-gated model by @yishuiliunian in #101
• fix(tui): prevent Shift+Enter inserting 'j' instead of newline by @yishuiliunian in #102
• fix(runtime): emit error event on silent SSE stream truncation by @yishuiliunian in #104
• feat(tui): add /skills slash command by @yishuiliunian in #103
• feat(tui): show session resume info on /exit by @yishuiliunian in #105
• feat(tui): /mcp slash command with server status page and reconnect by @yishuiliunian in #106
• feat(tui): event-driven background task sync and log viewer by @yishuiliunian in #108
• feat(tui): convert /skills to sub-page overlay by @yishuiliunian in #110
• feat(context): config hot-reload via ContextPipeline middleware by @yishuiliunian in #111
• fix(ipc): retry initialize() on transient failures by @yishuiliunian in #112
• fix: resolve UTF-8 boundary panics in session resume by @yishuiliunian in #113
• fix(sandbox): unrestrict signal, process-info and add mach-register by @yishuiliunian in #114
• feat(mcp): .mcp.json config support and disconnect action in /mcp page by @yishuiliunian in #115
• fix(context): prevent context overflow by clamping max_tokens before API call by @yishuiliunian in #116
• feat(tui): add Tasks panel with PanelProvider registry architecture by @yishuiliunian in #117

Full Changelog: v0.1.1...v0.2.0

v0.1.1

What's Changed

• fix(ci): patch rules_rust to fix Windows build PATH overflow by @yishuiliunian in #97
• feat(memory): upgrade sidebar memory agent to Knowledge Manager by @yishuiliunian in #98
• refactor: enforce child process stderr isolation to protect TUI by @yishuiliunian in #99
• feat(agent): depth-limited spawning, hub budget, and fork context by @yishuiliunian in #100

Full Changelog: v0.1.0...v0.1.1

v0.1.0

What's Changed

• fix: align macOS seatbelt sandbox with Codex whitelist policy by @yishuiliunian in #58
• fix: ESC interrupt lost + background agent results not delivered to root by @yishuiliunian in #59
• feat: centralized FocusMode state machine for TUI input dispatch by @yishuiliunian in #60
• feat: enhance Plan mode with 5-phase workflow, hard enforcement, and user approval by @yishuiliunian in #61
• feat: MetaHub distributed agent cluster + execution mode refactoring by @yishuiliunian in #62
• feat: enhance Agent tool call display with name, spinner, and live progress by @yishuiliunian in #63
• fix: sub-agent returning empty output when final LLM response has no text by @yishuiliunian in #64
• feat: background task architecture refactoring by @yishuiliunian in #65
• feat: enable topology overlay by default by @yishuiliunian in #66
• refactor: replace raw u64 timeouts with Duration for type-safe time handling by @yishuiliunian in #68
• feat: display full thinking content in TUI by @yishuiliunian in #67
• feat: add diff preview for MultiEdit, Write, and ApplyPatch in TUI by @yishuiliunian in #69
• refactor: simplify macOS Seatbelt sandbox from whitelists to blanket allows by @yishuiliunian in #70
• feat: add table cell borders and centralize markdown styles by @yishuiliunian in #71
• fix: resolve /exit 5-second delay by properly closing transport writer by @yishuiliunian in #72
• feat: agent-type-aware system prompt with fragment selection by @yishuiliunian in #73
• fix(skill): add conflict detection to github-pr by @yishuiliunian in #74
• feat: upgrade hook system with multi-executor types, structured output, and event telemetry by @yishuiliunian in #75
• fix(provider): resilient HTTP client with auto-rebuild on pool corruption by @yishuiliunian in #76
• feat: enhance prompt system with feature-conditional fragments and enriched tool descriptions by @yishuiliunian in #77
• feat: add session resume as agent-level capability with /resume command by @yishuiliunian in #78
• fix(tui): Up/Down keys navigate history instead of scrolling content by @yishuiliunian in #79
• feat(tui): add session picker for /resume + auto session titles by @yishuiliunian in #80
• feat(tui): add full-screen tabbed /status sub-page by @yishuiliunian in #81
• fix(runtime): detect SSE stream truncation and auto-continue from partial response by @yishuiliunian in #82
• refactor: deliver messages directly to agent mailbox, remove TUI inbox routing by @yishuiliunian in #84
• feat(sandbox): replace hard denies with permission-gated RequiresApproval by @yishuiliunian in #83
• fix(tui): decouple mouse scroll from input history via timing-based debounce by @yishuiliunian in #85
• feat(prompt): add concurrency discipline for sub-agent spawning by @yishuiliunian in #86
• feat(agent): depth-aware delegation + StartAgentParams refactor by @yishuiliunian in #87
• fix(tui): replace scroll debounce with batch detection and absolute position indexing by @yishuiliunian in #88
• fix(tui): replace alternate scroll heuristic with MouseCapture by @yishuiliunian in #89
• fix(provider): thinking-mode prefill rejection + error misclassification by @yishuiliunian in #90
• docs(readme): fix tool count, sandbox policies and expand feature coverage by @yishuiliunian in #92
• fix(hub): close transport after IO loop to prevent sub-agent process leak by @yishuiliunian in #93
• fix(config): flaky test due to LOOPAL_MODEL env var race by @yishuiliunian in #94
• feat(telemetry): OpenTelemetry integration with JSONL file export by @yishuiliunian in #95
• fix(runtime): decouple tool dispatch from permission to prevent duplicate tool_result by @yishuiliunian in #96

Full Changelog: v0.0.1-alpha.2...v0.1.0

v0.0.1-alpha.2

What's Changed

• refactor: agent state machine, deterministic result passing, overflow-to-file by @yishuiliunian in #57

Full Changelog: v0.0.1-alpha.1...v0.0.1-alpha.2

Full Changelog: v0.0.1-alpha.1...v0.0.1-alpha.2

v0.0.1-alpha.1

What's Changed

• feat: migrate build system from Cargo to Bazel by @yishuiliunian in #39
• chore: add Makefile with common development commands by @yishuiliunian in #40
• fix: convert Bash tool timeout from seconds to milliseconds & remove WebSearch by @yishuiliunian in #41
• fix: decouple mouse scroll from input history navigation by @yishuiliunian in #42
• feat: add --version flag and make install target by @yishuiliunian in #43
• refactor: introduce TimeoutSecs newtype to eliminate timeout unit ambiguity by @yishuiliunian in #44
• feat: task-based model routing + provider enrichment by @yishuiliunian in #45
• feat: add cron scheduler for timed prompt injection by @yishuiliunian in #46
• feat: ACP standard alignment + Hub architecture split by @yishuiliunian in #47
• refactor: remove misleading token metrics from TUI status bar by @yishuiliunian in #48
• refactor: simplify CLI — remove --hub, hide --serve by @yishuiliunian in #49
• feat: make /init agent-powered — analyze project before generating LOOPAL.md by @yishuiliunian in #50
• chore: add github-pr skill for automated PR workflow by @yishuiliunian in #51
• feat: collapse skill invocations in TUI chat view by @yishuiliunian in #52
• feat: agent view switching — TUI can observe and interact with any agent by @yishuiliunian in #53
• feat: add Auto permission mode with LLM-based tool classification by @yishuiliunian in #54
• refactor: decouple agent layer from UI — pure service model by @yishuiliunian in #55
• fix: remove duplicate root agent in topology overlay by @yishuiliunian in #56

Full Changelog: v0.0.1-alpha...v0.0.1-alpha.1

Full Changelog: v0.0.1-alpha...v0.0.1-alpha.1

v0.0.1-alpha

Loopal v0.0.1-alpha

First alpha release of Loopal — an AI coding agent with TUI.

Highlights

• Multi-agent architecture: Hub-only gateway with parent/child agent topology
• MCP support: rmcp SDK integration with full protocol support
• TUI: ratatui-based terminal UI with agent panel, model picker, topology overlay
• Built-in tools: Read, Write, Edit, Bash, Grep, Glob, WebFetch, and more
• LLM providers: Anthropic, OpenAI, Google, OpenAI-compatible
• Context management: Middleware pipeline with smart compaction
• Permission system: ReadOnly / Supervised / Dangerous tool levels

Recent changes

• /topology command: graphical Canvas-based agent topology overlay (#37)
• Hub-only gateway architecture with agent collaboration (#36)
• MCP migration to rmcp SDK (#35)
• Fix: release workflow runner for macOS x86_64 (#38)

Note

This is an early alpha release for testing and feedback. APIs and configuration formats may change.

What's Changed

• fix: /compact command provides TUI feedback via Compacted event by @yishuiliunian in #1
• feat: agent loop architecture enhancement with Runner decomposition and 7 new capabilities by @yishuiliunian in #2
• feat: server-side web search for all providers + OpenAI Responses API migration by @yishuiliunian in #4
• fix: replace Notify with watch channel for interrupt cancellation by @yishuiliunian in #3
• fix: enable bracketed paste mode to prevent multiline paste splitting by @yishuiliunian in #5
• fix: replace mouse capture with alternate scroll for terminal text selection by @yishuiliunian in #6
• feat: generalize server tool handling + redesign compaction architecture by @yishuiliunian in #7
• feat: prompt engine, TUI improvements & bug fixes by @yishuiliunian in #8
• refactor: delegate grep/glob to Backend with parallel search by @yishuiliunian in #9
• fix: add dedicated tool preference warnings to Bash description by @yishuiliunian in #10
• feat: budget-constrained ContextStore replaces raw Vec by @yishuiliunian in #11
• fix: update model params and sanitize orphaned server tool blocks by @yishuiliunian in #12
• fix: replace xterm alternate scroll with native mouse capture by @yishuiliunian in #13
• feat: integration test infrastructure & comprehensive test coverage by @yishuiliunian in #14
• feat: display actual server tool result content in TUI by @yishuiliunian in #16
• feat: replace repeated retry errors with in-place transient banner by @yishuiliunian in #15
• refactor: unify slash command system with CommandHandler trait + registry by @yishuiliunian in #17
• feat: multi-process architecture with TUI/Agent IPC isolation by @yishuiliunian in #18
• feat: incremental markdown rendering + table cell wrapping by @yishuiliunian in #19
• fix: unify context budget with model-aware context window by @yishuiliunian in #22
• feat: improve plan mode workflow and enforce tool usage over Bash by @yishuiliunian in #20
• fix: ESC interrupt not reaching agent during tool execution in IPC mode by @yishuiliunian in #21
• feat: show inline diff content in Edit tool TUI display by @yishuiliunian in #23
• fix: resolve IPC pipe deadlock + add multi-process observability logging by @yishuiliunian in #24
• refactor: replace dynamic worktree tools with static isolation model by @yishuiliunian in #25
• refactor: remove unused cost/price dead code by @yishuiliunian in #26
• feat: multi-agent IPC architecture + type-safe completion signal by @yishuiliunian in #27
• refactor: remove single-process mode, unify on multi-process IPC by @yishuiliunian in #28
• fix: resolve HubFrontend stale interrupt causing agent loop exit by @yishuiliunian in #29
• fix: restore Up/Down history navigation when agent is idle by @yishuiliunian in #30
• refactor: extract loopal-agent-hub crate from session by @yishuiliunian in #31
• fix: route server tool input deltas + clear stale sub-agent status by @yishuiliunian in #33
• chore: switch from MIT to proprietary commercial license by @yishuiliunian in #34
• feat: migrate MCP to rmcp SDK with full protocol support by @yishuiliunian in #35
• refactor: Hub-only gateway architecture + agent collaboration overhaul by @yishuiliunian in #36
• feat: add /topology command with Canvas-based agent graph overlay by @yishuiliunian in #37
• fix: use macos-14 runner for x86_64-apple-darwin release build by @yishuiliunian in #38

New Contributors

• @yishuiliunian made their first contribution in #1

Full Changelog: https://github.com/AgentsMesh/Loopal/commits/v0.0.1-alpha

Full Changelog: https://github.com/AgentsMesh/Loopal/commits/v0.0.1-alpha