All notable changes to Loom are documented in this file.

This changelog is generated directly from git commit history (non-merge commits) to keep entries accurate and auditable.

• Add DB migration 20260416_007_conversation_turn_metadata_v1 so thread user turns can durably persist explicit attachment/context metadata across resumes
• Add explicit conversation-turn attachment metadata plumbing for desktop thread file mentions and pasted-image attachments
• Refactor MCP + auth usability around a shared account authority: new MCP OAuth writes now use Loom secret refs with fingerprinted server identity, workspace-defined remote servers expose approval/trust state, and the desktop app now has a dedicated Integrations tab with management-grade MCP/account state plus guided account connect/routing actions

• Add DB migration 20260403_006_search_provider_state_v1 for authoritative auth-free search provider pacing/cooldown state
• Refactor web_search around shared SQLite-backed provider authority: DuckDuckGo is now the primary auth-free provider, provider pacing is coordinated across Loom runs, and DDG waits are skipped when they would violate the tool runtime budget
• Unify task/cowork data authority boundaries: canonical task snapshots now gate run-control writes, cowork semantics come from conversation_turns, transcript replay is journal-coverage-aware, and cowork session checkpoints record the turn boundary they cover
• Add DB migration 20260402_005_data_authority_unification for task snapshot freshness plus cowork checkpoint/journal coverage metadata
• Tighten cowork projection discipline: session checkpoints now derive counters from committed turns, non-semantic session metadata patches no longer rewrite checkpoint state, and loom db doctor warns on uncovered legacy transcript journals
• Refactor loomd desktop/runtime hot paths around SQLite: long-lived governed connections, governed durable-run lease writes, bounded durable event backpressure, direct live run/notification SSE delivery, and reduced frontend polling overlap
• Add desktop hot-path latency diagnostics and a synthetic active-run API smoke benchmark (scripts/active_run_latency_smoke.py)
• Add loomd sidecar entrypoint and runtime status contract for workspace-first desktop bootstrapping
• Add workspace registry, workspace settings, and conversation/run link schema with migration coverage
• Add workspace-first API surfaces: /runtime, /workspaces, /conversations, /runs, and grouped /settings
• Scaffold apps/desktop monorepo workspace with Tauri 2 + React + Vite overview shell

• 2026-03-16 24a6ea4 Fix tests for new streaming and cowork defaults
• 2026-03-16 e5c9905 Align default config with tuned local settings
• 2026-03-15 5b6696d feat: add /config runtime controls and live delegate timeout updates
• 2026-03-15 0d3c0b0 Handle adaptive replans for resumed verification runs

• 2026-03-08 f0c95f7 Align tier1 syntax hard-fail scope with retry-writable outputs
• 2026-03-08 0fa78e1 fix: enforce resilience kill-switch paths and parse retry routing
• 2026-03-08 2147e47 telemetry: add verification resilience reliability metrics
• 2026-03-08 a4dae41 refactor: add assertion envelope and profile-aware synthesis resilience
• 2026-03-08 a91ae16 refactor: add resilience decision matrix and profile classification
• 2026-03-08 73e3b43 Harden verification reliability and add resilience implementation plan
• 2026-03-08 388087d Mark command-wrapper tools as mutating for seal guard coverage
• 2026-03-08 9cdf44e Improve valuation verification reliability for precedent evidence
• 2026-03-08 9864f4b modified readme
• 2026-03-08 e03fd94 modified readme
• 2026-03-08 62ff36d Reseal artifacts after placeholder prepass mutations
• 2026-03-07 965099c Changelog: mark version standardization as 0.2.0 release

• Normalize package version authority to pyproject.toml and bump to 0.2.0
• Replace hardcoded runtime version literals with shared loom.__version__ usage
• Add version consistency guardrails (scripts/check_version_consistency.py, CI check, and tests)
• Update release-facing docs/examples for 0.2.0 (README.md, INSTALL.md, docs/tutorial.html)

• 675c4fe Initial commit
• ae23f7c Add project overview spec (00-PROJECT-OVERVIEW.md)
• 17acd74 Add project structure spec (01-PROJECT-STRUCTURE.md)
• 5f12e64 Add orchestrator loop spec (02-ORCHESTRATOR-LOOP.md)
• 3286a68 Add task state/memory and model router specs (03, 04)
• a051fb0 Add tool system spec (05-TOOL-SYSTEM.md)
• 03bb276 Add verification gates spec (06-VERIFICATION-GATES.md)
• bb5a7df Add API server spec (07-API-SERVER.md)
• cb6ed66 Add event system spec (08-EVENT-SYSTEM.md)
• 89b363d Add TUI client and agent integration specs (09, 10)
• f9a0cf3 Add workspace and file management spec (11-WORKSPACE-FILES.md)
• ee7d15a Add prompt architecture and error recovery specs (12, 13)
• 0fd04f3 Add human-in-the-loop and learning system specs (14, 15)
• c7f4cc2 Scaffold project: pyproject.toml, CLI, config, CI, test infra
• b562cf1 Implement task state and memory layer (Spec 03)
• e433dd3 Implement prompt template system (Spec 12)
• 53115f1 Implement model router with Ollama and OpenAI providers (Spec 04)
• 90b2a95 Implement tool system with registry, file ops, shell, search (Spec 05)
• 1741fae Implement workspace and file management (Spec 11)
• f21b625 Implement orchestrator loop with event bus and scheduler (Spec 02)
• b1a670f Implement API server with FastAPI endpoints and SSE streaming (Spec 07)
• b1f6e11 Add README, installation guide, and web-based tutorial
• 39b008c Fix runtime bugs: /config tier attribute and engine scratch_dir path

• 41845e8 Wire Phase 2 integrations: verification gates, changelog, memory extraction, event persistence, re-planning
• 24da424 Update docs for Phase 2: verification gates, event persistence, re-planning
• 1b4b3b0 Implement confidence scoring, approval gates, retry escalation, and webhook delivery (Specs 08, 13, 14)
• f97a5a5 Implement TUI client, MCP server, and learning system (Specs 09, 10, 15)
• 09794d3 Fix CI: use --extra dev for optional dependency installation
• 70ae27f Update README and INSTALL docs for all implemented features
• f269e31 Update tutorial with TUI, approval gates, MCP, and learning sections
• 5d1f29c Add full end-to-end integration test and agent connection documentation
• 5d8bc94 Add SubtaskRunner class for subtask execution encapsulation
• 0d61a93 Refactor orchestrator: extract SubtaskRunner, add parallel dispatch
• 396f524 Update docs and add CHANGELOG for SubtaskRunner and parallel dispatch
• c89ce04 Add implementation plan for bringing Loom up to snuff
• 8630e8d Implement phases 1, 2, and 5: new tools, streaming, error intelligence
• 0597406 Implement phases 3, 4, and 6: TUI enhancements, smarter planning, conversation mode
• b15311d Refactor tool registration to plugin auto-discovery model
• d7049c0 Update CHANGELOG with plugin auto-discovery entry
• f4dc27d Add gap analysis: Loom vs Claude Code cowork model
• 84b59fa Implement cowork mode: conversation-first interactive execution
• 620437d Update CHANGELOG and README with cowork mode, new tools, and Anthropic provider
• 926be5e Add web_search tool using DuckDuckGo (no API key required)
• 19cf277 Add per-tool-call approval system for cowork mode
• c0d9c9d Add streaming, task tracker, PDF/image support; close all gaps
• cf9cb55 Rewrite TUI to use CoworkSession directly (no server required)
• 112c73f Update all docs and tutorials to reflect cowork features
• 4a14ba3 Fix all ruff lint errors: unused imports, line length, naming conventions
• 83c6343 Move PLUGIN_PLAN.md to planning/refactors/
• 2b018d7 Add TUI refactor plan: multi-panel command center design
• 55f99fb Refactor TUI into multi-panel command center with dark theme
• ba37f64 Add comprehensive error handling across all system layers
• 6982e1c Clarify that cowork mode works with any model provider
• d5df627 Add conversation archive refactor plan for non-lossy cowork history
• 2a152be Add three-layer context awareness strategy to conversation archive plan
• 64427b9 Add cowork-to-task delegation as core component of conversation archive plan
• ba64a59 Implement non-lossy conversation history + task delegation for cowork mode
• 71bf5df Add mid-session switching between cowork sessions
• 0bbfb49 Fix 52 bugs, security flaws, and edge cases from comprehensive code review
• d3cf583 Fix 23 bugs from second comprehensive code review
• 77d7f5b Fix 11 bugs from rounds 3-5 code review + changelog + audit doc

• 8b6d5e4 Add comprehensive plan for general-purpose Loom expansion
• 5a64836 Redesign expansion plan around declarative process definitions
• 0d2c4d5 Add three rounds of critical review to process definition architecture
• 5ca225d Fix 15 lint errors: line length, unused imports, f-string and alias issues
• d34e4fd Implement process definition plugin architecture with extensible tool system
• 7f17236 Clean up built-in process definition YAML formatting
• 6ada59a Fix correctness bugs: type safety, double stat, unused --process flag
• 6caa980 Harden safety: exponent limits, file size checks, error logging
• 28cae9b Fix type: ignore comments in spreadsheet tool with assertions
• 9bfaa5e Add tests for safety guards and edge cases
• 70ecb71 Remove type: ignore in cycle detection with proper type narrowing
• 70ed852 Minor YAML formatting cleanup in builtin process definitions
• cd40f19 Update all documentation for process definition plugin architecture
• 1a2d639 Add process package installer with loom install/uninstall commands
• a3dad14 Add security review gate to process package installer
• 1dc26b5 Round 1: Fix correctness bugs across verification, installer, schema, CLI
• d04a7cf Round 2: Safety and edge case hardening
• 20217aa Round 3: API consistency fixes
• 93b8742 Round 4: Improve test coverage for verification, schema, and tool fixes
• 43403fb Add Google Analytics process package
• e1de19b Fix lint errors: import ordering and collections.abc
• ba2f17f Add fuzzy matching, batch edits, and diff display to edit_file
• a36c757 Code review fixes: bug fixes, TUI diff support, expanded tests
• ea59164 Fix lint: line length violations in file_ops and tests
• 7e0eda0 Rewrite README for clarity, accuracy, and the human reader
• bd5b344 Add multimodal content support: images, PDFs, thinking blocks
• b144a87 Wire multimodal content through TUI, API, and terminal display
• e66c3ff Fix multimodal integration gaps: OpenAI provider, deserialization, events
• 2e07188 Harden multimodal content: fix data loss, injection, leaks, validation
• 0690100 Fix TUI rendering bugs: center modals, escape markup, style diffs
• 47f4353 Unify TUI as the single interactive interface; loom launches it by default
• b7e4587 Update all docs and tutorial for unified TUI as default interface
• 0be941d Harden TUI: graceful DB fallback, broader error handling, ask_user in followups
• a7821ea Add sad-path tests for TUI cowork integration
• 376b051 Add first-run setup wizard and loom setup command
• fc59cd1 Move setup wizard into TUI as a modal screen
• e1d92e5 Fix TUI setup bugs, harden session logic, update docs
• 3fdf077 Lint cleanup: remove unused imports and variables
• 2064c25 Fix CI lint failures: remove f-string prefix and sort imports
• 2be0121 Add process package authoring guide for AI and human authors
• eeb27a7 Broaden README to reflect Loom as a general-purpose agent engine
• 9410b17 Expand default workspace analysis to scan for non-code documents
• 20ee5b9 Add Word (.docx) and PowerPoint (.pptx) document support
• f866705 Fix lint error and update docs for office document support
• 98afef2 Update model names in README tagline to Kimi, Minimax, GLM
• 6e8b5d8 Replace remaining Qwen references with MiniMax in README
• 8e42288 Fix MiniMax model identifiers in README config examples
• 22a1c83 Use kimi-k2.5 as primary model in README config example

• da45203 Move PLAN.md into planning/refactors directory
• 6af21c4 Add automatic behavioral reflection to ALM (Spec 16)
• 13502e2 Add spec to drop regex from codebase in favor of structural alternatives
• 1830758 Rewrite DROP-REGEX spec to scope only to learning/reflection.py
• 8d3f0c6 Redesign ALM spec around task completion gap analysis
• 83ec647 Replace regex reflection engine with task completion gap analysis
• 832bf93 Add /learned command for reviewing and deleting learned patterns
• f50de9e Update docs and specs for ALM and /learned command
• ab46d78 Update tutorial.html for ALM and /learned command
• faef639 Add help and quit to command palette; update README intro
• fc38c86 Redraft README intro sections: Claude cowork framing and MCP mention
• 489e694 Add foundation refactor plan from full codebase audit
• f7be8ce Add tree-sitter integration plan from design discussion
• 03490ce Implement 6-phase foundation refactor across 22 files
• 71cca12 Fix linter errors: import ordering and line lengths
• ceeaa59 Fix duplicate tool registration crash on TUI startup
• e62c125 Rename loom.toml to loom.toml.example, gitignore loom.toml
• 2316bd6 Fix setup wizard not accepting keystrokes
• edc7d2d adding code review
• 9fbe157 Add code review plan from main branch
• 026b94f tui fixes and review
• 4d488ad Fix 14 code review issues (P0-P2) with tests
• 0894a4c Add TUI review document from main branch
• 9e6a8cf Fix 8 TUI review findings (P0-P2) from 2026-02-16-TUI-REVIEW
• e1c694d Update docs to reflect current codebase state
• 0e727be Add tree-sitter integration for code analysis and structural editing
• b80f32a Fix review findings: byte/char offsets, nested defs, Rust impl, naming

• 1f0f2ff Update docs for tree-sitter integration and current stats

• a44cdcf Fix API process isolation, reset TUI file panel, and add tree-sitter CI gate
• 770152d Harden weak tests to reduce false positives
• 6404dd6 code-review docs and uv.lock
• ab0de15 Harden TUI setup flow and add process/package refactor plan
• 39750b4 Propagate active process through TUI sessions and delegation
• fdacb0b Enforce process required tools in TUI and orchestrator
• 35eb106 Honor process.default for run and TUI launch
• 2256c7f Add in-session /process controls and process status display
• c72f542 Copy full process package assets with safe installer filtering
• 670f5b7 Enforce strict process phase blueprints in planning
• d37618d Fix package install docs for Google Analytics process
• 7c4894d Add process controls to the TUI command palette
• d4377f4 Reject process tasks with missing required tools at API create
• 3e1d2db Document in-session process controls in README
• 3c5e99b Apply process.default in API task creation path
• 4abb24c Enforce process runtime semantics for critical and synthesis phases
• 8eb1ae4 Close process/package plan: isolated deps, collisions, docs
• 3226783 Harden web tools and scope process deliverable verification

• f174ac0 feat: add process testing framework and MCP tooling improvements
• 559ee89 fix: apply configured api keys in openai-compatible and ollama providers
• 334f03e refactor: broaden cowork system prompt beyond coding-only framing
• d7da827 Improve process orchestration UX, resiliency, and docs parity
• 37e968e Add rich TUI file viewer renderers and learned-pattern cleanup
• e273c04 Close file viewer modal on backdrop click
• 663ebb3 Implement MCP config management across CLI, TUI, and runtime
• 3fd3d8e Refine MCP OAuth plan and fix TUI input/hint layout
• c0dd66b Improve process run live feedback and process-use UX
• a66447e Expand Ctrl+P palette coverage and command prefills
• 85d1efb Harden process runs, TUI UX, and provider message compatibility
• 8d71f5d Harden verifier JSON parsing for critical-path retries
• e858f11 Harden OpenAI payloads against empty assistant content
• 92e4d4e Add market-research process and harden TUI process UX
• c815333 Replace hard truncation with semantic compaction

• 1a9841e Harden process verification stack and tighten delegate logging

• e1dc880 massive processing refactor. v2 of process definition.
• 84c5798 TUI hierarchy fixes
• ba3c7a4 Fixing some process issues, TUI issues, and more.
• bf26548 fixing a small bug in testing and reworked the process progress box in the TUI
• 028b4e5 small ui bug
• 21cfe1d doc updates

• 28b4caa Fix TUI mouse interaction startup flag
• ec0c240 Simplify MCP/auth UX and prune routing complexity

• 4f42efe Tune compactor budgeting and docs defaults
• f1d80d2 Align CLI run process resolution with TUI flow
• c883f64 Stabilize event persistence assertions in integration tests
• 2d28ae3 Harden replanning with version fencing and strict ID continuity

• 1fe5202 docs: align model role split examples and revalidate dead-code plan
• ccb07fb feat: add artifact ingest handling and telemetry transparency controls
• b92d1b6 Implement telemetry transparency events and default-on rollout
• d88f5f2 Harden contradiction scan coverage gating
• 4fba048 Avoid hard import dependency for contradiction event type
• d082fdc Implement keyless research tool suite
• a2d66d2 doc updates, process updates, retry architecutre changes
• c177ba6 adding potential web frontend
• 1f7f7e9 Export evidence ledger CSV during wrap-up

• 79295b3 Harden economic provider JSON payload parsing
• 9aa8371 Honor read_roots for spreadsheet read operations
• 63eaa9c Add /run goal-file input support and docs
• 291437f Add humanize_writing tool with deterministic scoring and tests
• 6e43efe Make map-competition non-blocking for unverified competitor entries

• 3318acd Add keyless investment tool suite and FRED economic provider support
• 1d5524a Add TUI input history navigation with resume restore
• 3b24715 Fix portfolio optimizer event-loop freeze in TUI runs
• a0e495f Add keyless market-signal tool suite
• 6914233 Fix synthesis deadlocks and add stalled-run safeguards
• d2dd913 Fix deterministic synthesis contract responses
• 84e95f5 docs: align tutorial intro with harness positioning
• 0e22f88 Make missing-data remediation non-fatal in marketing strategy

• c5880f7 docs: use uv run loom in README commands
• 3ebb559 adding technical design doc
• e088aa9 Implement reliability hardening and refresh technical design

• 8fb9b59 docs: update README.md
• 551670d docs: add README screenshot
• 0de3623 Implement resource-first auth lifecycle, migration tooling, and UX hardening
• d133ef2 Harden process prompts and shell safety for segmentation runs
• f3747a4 Improve cowork turn telemetry and render chat as Markdown
• e98d3d5 Reduce startup and interaction latency via async MCP warmup
• d1dfa4f Add hybrid tool entrypoints and follow-up config/auth updates
• 338f427 Harden auth profile selection and tool inventory coverage

• dc3afe9 cowork: compact context via recall index and add turn telemetry
• 55156ee Improve hybrid list_tools flow and enforce scoped schema lookup
• b94f013 Fix cowork prompt JSON examples in f-string
• c2e61ff Repair dangling tool-call chains in cowork context
• 4536fb3 Add model-planned failure remediation with bounded metadata context
• 8ef408a Add cowork tool-loop convergence guard and context-window fix
• f1b3064 Add structured match and file counts to ripgrep_search
• f23d72d Implement chat history replay refactor for TUI session resume
• 3a743cf Harden slash command hints with full scrollable catalog
• e994e59 Fix duplicate chat panel rows during live turns
• 0883857 Implement cowork delegate progress UX and realtime TUI refresh
• 44db5b9 Fix lint line-length in delegate progress widget
• 1184a00 Harden /run launch liveness and in-place stage heartbeat UX
• 2edf67e Harden process-run cancel flow and fix queued Ctrl+W hangs
• b5df6ce Move technical design doc to docs with updated date
• 5cda3c2 changing image
• 1a139bf Add live GA4 API retrieval with auth to google-analytics package
• 1e59c87 Fix ga_live_api tests leaking global tool registrations
• ab377f5 Move GA live API tests into package-local test directory

• c4630d8 Improve auth discovery scope and preserve selected mode in /auth
• d6e05ef tui: remove /process mode and simplify /run palette flows
• 38f57c9 mcp: add oauth token runtime and manager support
• 486a6b6 Harden MCP/Auth management and tab UX
• c876a05 Add header activity indicator and refine active animation
• 0ad619a Add rich /model details and new /models TUI command
• 1738235 Refine header activity indicator rendering behavior
• 17d9b20 Align outputs with replans via phase IDs; remove deprecated auth-path noise
• 48bda35 tui: align sidebar progress icons in dedicated column
• 9d1a0f4 Fix MCP stale alias reload and auth-check explicit scoping
• e0d72f6 Implement hybrid cowork chat stop control and hardening
• 227c1e7 Implement cowork steering queue controls and execution fixes
• fa6035e Fix remote MCP OAuth and streamable MCP tool discovery
• 9c20a61 Enable clickable markdown links with URL tooltip in cowork chat
• 7c236df auth: harden mcp oauth boundary and reuse compatible drafts
• 72bb236 tui: make auth manager sync explicit and refresh read-only
• 951db4c Retheme cowork markdown to replace magenta defaults
• e3e2ad4 auth: classify audit history separately from active orphan findings
• c9fbfc1 auth: add /auth oauth lifecycle commands and tui actions

• 241e47c Complete auth cleanup phases: OAuth parity, TUI mode UX, and token handling
• e8a2d6b docs: sync endpoint count and current project stats
• 5dac6ba changing image
• e2dba38 tests: remove credential-style URI fixture
• ae69a68 ci: keep live canary config out of uploaded artifacts
• 7416521 Improve tooling UX and provider-specific coding agent integrations
• 385c66a Guard on_key screen lookup when no screen stack is mounted

• ed92c32 Add phase-level iteration loop runtime, gates, persistence, and docs
• 42b5e08 Improve process run controls, auth preflight, and workspace selection
• a30398d Fix chat scroll lock and tighten activity pane spacing
• 5c36ed6 feat(cowork): harden long-session context recall and indexing

• 3456125 Harden ask_user runtime flow and execution-surface gating
• ed2a6bc Align delegate_task TUI approval default with /run
• f3c68ca docs: remove stale LOC remark from README
• 1ce7cca adding trufflehog precommit hook

• 0813474 Improve command palette, landing flow, and close-tab safety
• 031a631 Add loom2 hero image to README
• 3dd0b07 feat(validity): enforce claim-level synthesis gates and evidence lineage
• 6b72edf Enforce evidence-backed edits for sealed verified artifacts
• bfdd915 Enable ask_user interactive defaults
• 935e21d Default process test execution to temp workspace

• ca97094 Refine ask_user UX and raise per-subtask cap default
• b75d87a Split placeholder prepass from confirm-or-prune resolution
• fecb07a Harden SQLite migrations and startup persistence semantics
• 8aaefc4 Implement telemetry coverage and auditability hardening
• 03a16be adding refactor plans and updates to technical doc
• c42ebbd Relink migration diagnostics to event type constants
• 0d77f06 Fix verification guard wording and token redaction
• d8df362 Harden fan-in output coordination and transactional publish flows
• 0d70627 Implement runtime telemetry verbosity modes and controls

• a7fef18 Refactor core engine modules into package-based subsystems
• 48ad5ec refactor(tui): split app into package modules with css file
• cc90b6a refactor(cli): split main into modular cli package
• 8ac0558 docs(plan): update main entrypoint split plan
• 0c86629 Split monolithic TUI tests into tests/tui modules
• ab1647b Split orchestrator tests into package and add legacy path guard
• d8ab27d Unify sealed artifact mutation policy and sanitize docs paths