跳轉到

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, with entries grouped by release date.

[v0.5.0] - 2026-04-11

Changed

  • Refactored import-path migration scaffolding so mailslide is the canonical package surface while outlook_mail_extractor remains a warning-gated compatibility path.
  • Added deprecation schedule note: legacy outlook_mail_extractor import path is planned for removal in the next major release.
  • Refactored configuration validation/loading flow to use typed models (AppConfig/JobConfig) and a repository-based migration+backup write path.
  • Refactored plugin execution contract to require structured PluginExecutionResult returns and reject legacy boolean plugin results.
  • Added migration mapping notes for strict plugin-result contracts and modularized CLI boundaries.

Added

  • Added compatibility and regression coverage for import migration behavior in tests/test_import_compat.py.
  • Added typed config model and repository regression coverage in tests/test_config_models.py and tests/test_config_repository.py.
  • Added plugin-runner regression coverage for strict result-type enforcement and contract error mapping in tests/test_plugin_runner.py.

[v0.4.2] - 2026-04-03

Changed

  • Changed Main Configuration Job actions (Add/Edit/Remove) to persist immediately to config/config.yaml using the same sanitize/validation/write pipeline as the main Save path.
  • Changed Add/Edit modal save flow to persist before dismissing; modal now stays open on save failure and allows retry.
  • Changed Job action behavior to block when YAML editor text has unsaved changes, with explicit guidance to Save or Reload first.
  • Changed Home-tab synchronization behavior so entering Home now auto-refreshes the Jobs area from persisted config.yaml, removing the need for manual Refresh after Config job edits.
  • Changed refresh orchestration to use one shared Home reload path for both manual and automatic refresh triggers, keeping render semantics consistent.
  • Changed Main/LLM config tabs to modal-only editing surfaces by removing raw YAML text-area display from both pages.
  • Changed Main tab action model to remove YAML-oriented Validate/Save/Reload flow and expose a dedicated General Settings modal path instead.
  • Changed Main reset confirmation copy to explicitly warn that reset is a full overwrite and may remove custom unknown keys.
  • Changed Main General Settings schema mapping to enforce app-owned i18n metadata for fixed fields (body_max_length, llm_mode, plugin_modules), keeping key-driven rendering and English baseline labels even when _ui.fields supplies custom plain labels.
  • Changed Home execution controls to enforce a single dominant action by state (Run primary when idle; Stop emphasized when running) while keeping Refresh secondary.
  • Changed plugin config editor modal layout so action controls remain visible during long-form scrolling, and standardized action semantics (Save primary, Validate neutral-secondary, Cancel neutral).
  • Changed LLM connection test UX to show session-scoped inline diagnostics near Test Connection (latest outcome, last success timestamp, latency, and latest issue).
  • Changed footer shortcut hints to context-filtered rendering by active tab/modal while preserving existing keybinding behavior.
  • Changed dark-theme focus styling for Home table focus/cursor state, plugin-editor input focus, and Config tab active/focus states for clearer keyboard navigation.

Added

  • Added saving-state guardrails to prevent duplicate submit during modal save and Job remove operations.
  • Added/updated regression coverage for immediate persistence, dirty-editor blocking, modal failure retry behavior, and remove failure state handling in tests/test_main_config_editor.py and tests/test_add_job_modal.py.
  • Added new i18n messages for dirty-editor Job-action blocking and modal save-failure fallback in outlook_mail_extractor/locales/en-US.yaml and outlook_mail_extractor/locales/zh-TW.yaml.
  • Added deferred/coalesced Home auto-refresh handling while job execution is active, with one-shot catch-up after execution returns idle.
  • Added app-level Home tab activation wiring plus regression coverage for keyboard and tab-switch trigger paths in tests/test_app_home_tab_sync.py.
  • Added dedicated Home auto-refresh regression coverage (idle, running defer, inactive catch-up, and failure warning paths) in tests/test_home_auto_refresh.py.
  • Added new Home auto-refresh failure i18n warning keys in both locale files with manual Refresh retry guidance.
  • Added Main General Settings modal flow for top-level keys (body_max_length, llm_mode, plugin_modules) with validation-first immediate persistence.
  • Added save-attempt callback support to PluginConfigEditorModal so modal saves can stay open/retry on persistence failure.
  • Added/updated regression coverage for modal-only Main/LLM editing behavior in tests/test_main_config_editor.py and tests/test_llm_config_editor.py.
  • Added regression coverage for General Settings locale rendering (en-US/zh-TW) and missing-locale English fallback behavior in tests/test_main_config_editor.py.
  • Added LLM inline-status i18n keys in both locale files for outcome state, timestamp formatting, latency display, and unexpected-response details.
  • Added regression coverage for Home dominant-action state transitions, footer visible-action filtering by context, and inline LLM status rendering updates in tests/test_home_auto_refresh.py, tests/test_app_home_tab_sync.py, and tests/test_llm_config_editor.py.
  • Added a plugin-editor schema-action regression test for validate/save action availability in tests/test_plugin_config_editor.py.

Fixed

  • Fixed the misleading "saved but not yet persisted" Job-edit experience where users had to click Save again in the main settings pane for changes to actually take effect.
  • Fixed stale Home Jobs state after Config job mutations by auto-syncing on Home tab entry and catch-up after deferred refresh conditions.
  • Fixed Main modal save behavior to preserve unknown/unmanaged config keys while mutating only modal-owned fields.
  • Fixed mixed-language General Settings modal labels when schema entries omitted label_key or used hardcoded labels, by ensuring fixed keys render through stable translation keys with en-US fallback.
  • Fixed a ContextualFooter runtime crash on app startup (ReactiveError: Footer is not defined on OutlookMailExtractor) by correcting footer key compact-state binding.
  • Fixed footer-context resolution during modal screens (for example quit confirmation) to avoid NoMatches crashes when top-level TabbedContent is temporarily unavailable.

[v0.4.1] - 2026-04-02

Changed

  • Promoted v0.4.1rc7 to stable v0.4.1.
  • Included the full v0.4.1rc5 to v0.4.1rc7 hardening set: download_attachments recall/stability updates, runtime plugin-option UX improvements, and release-gate attachment recall validation.

Fixed

  • Fixed mixed-index install drift by pinning runtime and LLM-extra httpx to 0.28.1, preventing accidental 1.0.dev* selection that breaks LLMClient initialization.

[v0.4.1rc7] - 2026-04-02

Changed

  • Pinned runtime and LLM-extra httpx dependency policy to ==0.28.1 to prevent resolver drift to pre-release 1.0.dev* builds.
  • Updated release metadata surfaces to v0.4.1rc7.

Fixed

  • Fixed uv tool + TestPyPI mixed-index installs that could pick httpx 1.0.dev3 under --index-strategy unsafe-best-match, which breaks LLMClient initialization (Client.__init__(base_url=...)).

[v0.4.1rc6] - 2026-04-02

Changed

  • Changed Main Configuration job editor plugin options to resolve from runtime plugin discovery (list_plugins) instead of static schema-only options, with normalized/deduplicated sorting.
  • Changed job-edit modal behavior to preserve previously configured but currently unavailable plugins, so existing job configs remain editable without silently dropping plugin IDs.
  • Changed Add Job plugin labels to mark unavailable plugins explicitly in the selection UI.
  • Updated .gitignore to ignore .context/ runtime workspace artifacts.

Added

  • Added i18n keys for unavailable plugin labels in both en-US and zh-TW locale files.
  • Added regression tests for runtime plugin-option resolution, unavailable-plugin preservation, and unavailable-plugin label rendering in add/edit job flows.

[v0.4.1rc5] - 2026-04-02

This RC consolidates all changes from v0.4.1rc1 to v0.4.1rc4 and adds final hardening for attachment-recall reliability.

Added

  • Added the non-LLM download_attachments plugin end-to-end (plugin registration, config samples, deterministic path planner utilities, startup validation, and regression coverage) as introduced across the early v0.4.1rc1/v0.4.1rc2 line.
  • Added attachment-recall release evidence assets under docs/releases/attachment-recall/ and a dedicated validator script scripts/validate_attachment_recall_evidence.py.
  • Added publish workflow enforcement so release validation now checks both dependency RC evidence and attachment-recall evidence before publishing.

Changed

  • Changed inline-attachment classification to a completeness-first policy: skip only on high-confidence inline signals (2-of-3 strong signals), with fallback-download counters for partial metadata cases.
  • Changed plugin execution details contract to deterministic machine-readable status/code output with canonical details.emails[], saved_relative_paths, saved_files, skipped_inline_reasons, and failed_files payloads.
  • Changed attachment path handling to stricter deterministic behavior for invalid names, collision indexing, and full-path budgeting (including startup viability checks shared with runtime rules).
  • Changed startup/runtime failure taxonomy for download_attachments to stable startup_* and runtime_* code families, with deterministic precedence for run-level codes.
  • Changed RC evidence tooling from the v0.4.1rc3 stream to enforce filled templates and preserve automated scaffold generation for X.Y.ZrcN bumps.

Fixed

  • Fixed Windows absolute-path normalization/root-availability checks for download_attachments startup validation (from v0.4.1rc4) to avoid false negatives on non-Windows CI runners.

[v0.4.1rc4] - 2026-04-01

Changed

  • Fixed download_attachments startup-path validation and path normalization to treat Windows absolute paths consistently across non-Windows CI runners, avoiding false negatives in root-availability checks.
  • Added regression coverage in tests/test_job_execution_service.py for Windows absolute path normalization and unavailable-drive validation behavior.

[v0.4.1rc3] - 2026-04-01

Added

  • Added RC evidence scaffold generation in scripts/update_version.ps1 for X.Y.ZrcN bumps, creating docs/releases/evidence/X.Y.Z-rcN.md from template when missing.
  • Added regression coverage in tests/test_update_version_script.py to ensure RC evidence files are auto-created for RC version bumps.

Changed

  • Updated RC evidence validation in scripts/validate_rc_evidence.py to fail when template placeholders remain unfilled.
  • Updated release-evidence validator tests in tests/test_release_evidence_validation.py to cover unfilled-template rejection behavior.

[v0.4.1rc2] - 2026-04-01

Added

  • Added a new non-LLM plugin download_attachments that saves regular Outlook attachments to deterministic per-job folders and returns structured per-email/per-file execution details.
  • Added deterministic attachment path utilities (download_attachments_paths) for job-folder key generation, Windows-safe filename sanitization, case-insensitive collision handling, and fixed full-path budgeting.
  • Added startup validation for download_attachments output_dir in JobExecutionService so enabled jobs fail fast on missing/unresolvable path configuration.
  • Added new plugin sample config files for download_attachments under both runtime and packaged sample paths, and exposed it in config sample plugin options.
  • Added dedicated regression coverage for attachment path planning, plugin behavior, adapter metadata extraction, and startup validation (tests/test_download_attachments_paths.py, tests/test_download_attachments_plugin.py, tests/test_job_execution_service.py).

Changed

  • Extended MailActionPort with typed attachment operations (list_attachments, save_attachment) and added AttachmentDescriptor to keep plugin logic decoupled from Outlook COM internals.
  • Updated OutlookMailActionAdapter to surface attachment metadata (including inline hints/content-id) and support saving attachments through the action-port boundary.
  • Updated plugin job lifecycle context to include job_name during begin_job, enabling deterministic per-job output path derivation.
  • Updated existing plugin test fakes to implement the expanded action-port protocol.

[v0.4.0] - 2026-03-30

Changed

  • Promoted v0.4.0rc2 to stable v0.4.0 with the same dependency-guard, release-gate, and compatibility hardening introduced during RC validation.

Fixed

  • Fixed scripts/update_version.ps1 version-line matching so release bumps work consistently across LF and CRLF files on Windows.

Added

  • Added CRLF coverage in tests/test_update_version_script.py to prevent regression in automated release version updates.

[v0.4.0rc2] - 2026-03-30

Changed

  • Refactored EmailProcessor internals in outlook_mail_extractor/core.py to keep orchestration behavior stable while delegating message collection, email extraction, and job-metric aggregation to focused services.
  • Refactored plugin editor modal internals in outlook_mail_extractor/screens/modals/plugin_config_editor.py to separate payload/state/view concerns while preserving existing modal contracts and prompt-profile rename sync behavior.
  • Updated TUI language persistence in outlook_mail_extractor/tui.py to reuse shared YAML write/backup flow (write_yaml_with_backup) instead of inline file-write logic.
  • Updated plugin config loading in target plugins (calendar, event_table, summary_file, move, category) to use shared base-helper merge logic for common fields.
  • Updated release metadata surfaces to v0.4.0rc2 and added About-screen website rendering in outlook_mail_extractor/screens/about.py with i18n coverage in tests/test_i18n.py.
  • Updated release install guidance in README.md and README.zh-TW.md to document RC/TestPyPI compatibility policy and evidence location.
  • Updated packaging metadata to define canonical LLM policy in pyproject.toml (project.optional-dependencies.llm -> httpx<1) and synchronized uv.lock extras metadata.
  • Updated scripts/update_version.ps1 changelog insertion logic to be newline-safe and resilient when anchor text formatting changes.
  • Updated .gitignore to ignore docs/brainstorms planning artifacts.

Added

  • Added service extraction modules:
  • outlook_mail_extractor/services/email_extraction_service.py
  • outlook_mail_extractor/services/message_collector.py
  • outlook_mail_extractor/services/job_metrics_collector.py
  • Added plugin editor helper modules:
  • outlook_mail_extractor/screens/modals/plugin_editor_payload.py
  • outlook_mail_extractor/screens/modals/plugin_editor_state.py
  • outlook_mail_extractor/screens/modals/plugin_editor_view.py
  • Added/expanded characterization and regression coverage for refactor safety in tests/test_core_high_risk.py, tests/test_job_execution_service.py, and tests/test_plugin_config_editor.py.
  • Added a new compound learning document under docs/solutions/best-practices/service-orchestration-and-plugin-editor-decomposition-2026-03-29.md.
  • Added a runtime LLM dependency guard service (outlook_mail_extractor/services/dependency_guard.py) with a shared machine contract (outlook_mail_extractor/contracts/dependency_guard.py) for CLI/TUI failure signaling (12 / DEPENDENCY_GUARD_FAILED / dependency_guard_failed).
  • Added release guardrails for dependency-policy drift and RC evidence validation via scripts/check_llm_dependency_contract.py, scripts/validate_rc_evidence.py, and evidence templates under docs/releases/evidence/.
  • Added dedicated regression coverage for dependency-guard and release-gate behavior in tests/test_cli_dependency_guard.py, tests/test_home_dependency_guard.py, tests/test_dependency_guard_service.py, tests/test_llm_dependency_contract_drift.py, and tests/test_release_evidence_validation.py.

Fixed

  • Fixed a plugin-editor refactor regression risk by restoring constructor initialization to flow through modal hook/wrapper methods, avoiding direct helper-call bypass for extension/override-safe behavior.
  • Fixed compatibility exports for source launcher and migration package re-exports (app.py, mailslide/__init__.py) to preserve expected symbols during lint/packaging checks.
  • Fixed late TestPyPI LLM execution crashes by failing fast before processing loops when dependency compatibility policy is incompatible or unevaluable.
  • Fixed RC promotion traceability by linking release validation evidence to docs/releases/evidence/0.4.0-rc2.md.
  • Fixed a guard-ordering edge case by enforcing dependency preflight before LLM client initialization, preserving contract-stable failure signaling.
  • Fixed a prompt-profile override edge case so jobs that become LLM-required at runtime still trigger dependency-guard preflight.

Removed

  • Removed legacy release-note helper github_action_pypi_release.md after moving RC validation workflow guidance into CI gates, checklist flow, and evidence docs.

[v0.3.10] - 2026-03-29

Changed

  • Changed Home-tab stop behavior to use cooperative cancellation wiring end-to-end, propagating Textual worker cancellation signals from HomeScreen into job orchestration (JobExecutionService), per-mail processing (EmailProcessor), and LLM/plugin dispatch checkpoints.
  • Changed job/plugin lifecycle handling to preserve safe plugin finalization (end_job) even when cancellation occurs mid-run, while stopping new jobs/emails/plugins from starting after a stop request.
  • Changed Main Configuration job modal (新增 Job/修改 Job) to expose per-plugin prompt-profile inputs, so users can set plugin_prompt_profiles directly in the form instead of editing YAML manually.
  • Changed Plugin Configuration prompt-profile editor to allow renaming profile key values directly in the modal, with validation for non-empty and unique keys.
  • Changed plugin-save flow to auto-sync renamed prompt-profile references in config/config.yaml (jobs[].plugin_prompt_profiles) for the edited plugin, including backup-safe writes.
  • Changed Main Configuration tab to include a Reload action button that re-reads config/config.yaml from disk on demand (with schema fallback injection for older _ui samples that do not define the button yet).

Fixed

  • Fixed Home-tab ⏹️ 終止 / Stop button behavior where pressing stop only cancelled the UI worker but did not actually interrupt background execution; stop requests now halt execution at the next safe checkpoint (before next job, next mail, next plugin, and before each LLM call).
  • Fixed prompt-profile rename propagation reliability by adding plugin-side rename inference fallback and plugin-name callback fallback, so job plugin_prompt_profiles references are still synchronized even when explicit rename metadata is missing.
  • Fixed Linux/macOS CI import failures by guarding Outlook COM imports in core.py; non-Windows environments now raise a clear runtime OutlookConnectionError only when Outlook connection is attempted, instead of failing at module import time.

Added

  • Added cancellation regression coverage in tests/test_job_execution_service.py, tests/test_core_high_risk.py, and tests/test_llm_dispatcher.py to verify stop requests prevent continued processing beyond the current safe step.
  • Added modal/i18n coverage for per-plugin profile assignment in jobs, including tests/test_add_job_modal.py assertions for selected-plugin filtering and new locale strings in outlook_mail_extractor/locales/zh-TW.yaml and outlook_mail_extractor/locales/en-US.yaml.
  • Added prompt-profile rename + job-reference sync coverage in tests/test_plugin_config_editor.py, plus new plugin-editor/plugin-tab i18n messages for rename validation and sync notifications.

[v0.3.9] - 2026-03-28

Changed

  • Updated the LLM Configuration tab UI (outlook_mail_extractor/screens/config/llm_tab.py) to show an in-app AI mail-reading privacy warning and disclaimer directly below the LLM settings actions.
  • Added localized warning/disclaimer copy for both zh-TW and en-US in outlook_mail_extractor/locales/zh-TW.yaml and outlook_mail_extractor/locales/en-US.yaml so the legal/privacy notice follows the selected UI language.
  • Updated git ignore/tracking hygiene by untracking local release helper scripts (scripts/release_pypi.ps1, scripts/sync_guides.ps1) and adding them to .gitignore.
  • Updated event_table Excel writing (outlook_mail_extractor/plugins/event_table.py) to retry when files are locked by Excel, with configurable retry count/delay (excel_write_retries, excel_write_retry_delay_seconds) and explicit lock warning logs during retries.
  • Updated event_table config samples (config/plugins/event_table.yaml.sample, outlook_mail_extractor/resources/config_samples/plugins/event_table.yaml.sample) to expose the new Excel lock-retry settings in both runtime config and _ui schema.

Fixed

  • Fixed event_table lock contention handling so locked .xlsx writes now return retriable_failed with code="excel_file_locked" and a clear actionable message instead of a generic unexpected error.
  • Fixed malformed uv.lock content that broke uv run/test execution by restoring a valid [[package]] entry for mailslide.

[v0.3.8] - 2026-03-27

Added

  • Added terminal window title handling via outlook_mail_extractor/terminal_title.py, with Windows Console API support (SetConsoleTitleW) and ANSI fallback for TTY terminals.
  • Added config-driven terminal title override via terminal_title in config/config.yaml (default: Mailslide) for both CLI (mailslide) and TUI (mailslide-tui) startup flows.
  • Added terminal-title regression tests in tests/test_terminal_title.py covering Windows, ANSI, and config fallback behavior.

Changed

  • Updated config samples (config/config.yaml.sample, outlook_mail_extractor/resources/config_samples/config.yaml.sample) to document the new optional terminal_title key.

[v0.3.7] - 2026-03-27

Added

  • Added packaged Guide documentation resources under outlook_mail_extractor/resources/docs/ (GUIDE.md, GUIDE.en.md) so installed/tool environments can render Guide content without repository-local markdown files.
  • Added scripts/sync_guides.ps1 to sync/check root GUIDE*.md files against packaged docs resources, with -CheckOnly mode for CI/release validation.
  • Added Guide loading regression coverage in tests/test_usage_screen.py for packaged-resource fallbacks in both en-US and zh-TW locales.

Changed

  • Changed Guide tab content loading (outlook_mail_extractor/screens/usage.py) to fall back to packaged docs via importlib.resources when filesystem Guide/README candidates are unavailable.
  • Changed package data configuration (pyproject.toml) to include resources/docs/*.md in built distributions.
  • Changed PyPI release preflight (scripts/release_pypi.ps1) to enforce Guide resource sync before tests/build.

[v0.3.6] - 2026-03-27

Added

  • Added a Home-tab run-status ASCII animation above the jobs table (outlook_mail_extractor/screens/home.py) with a two-frame loop that switches every 0.5 seconds while jobs are executing.
  • Added packaged config-template resources under outlook_mail_extractor/resources/config_samples/** and runtime seeding utilities (config_templates.py) so uv tool installs can initialize config files without relying on repository-local config/ files.
  • Added a dedicated TUI entrypoint module (outlook_mail_extractor/tui.py) and script command mailslide-tui for tool-install execution (uv tool install mailslide + mailslide-tui).
  • Added release automation/docs artifacts for PyPI-first distribution, including scripts/release_pypi.ps1, tasks/release_checklist_pypi.md, and MANIFEST.in (prune tests) to keep source distributions lean.
  • Added job-execution path-normalization tests and service coverage (tests/test_job_execution_service.py) for plugin output-path resolution behavior.
  • Added plugin-editor path-resolution tests in tests/test_plugin_config_editor.py to cover absolute-path rendering for path fields with/without runtime base directory context.
  • Added Windows DPAPI-backed secret storage for llm-config API keys via outlook_mail_extractor/secrets.py, writing encrypted key material to config/llm-api-key.bin instead of plaintext YAML.
  • Added PyPI-based app update checks (outlook_mail_extractor/services/update_check.py) with startup delayed-check behavior (20s), plus About-tab update status display and a manual "Check Updates" action.
  • Added config-schema migration scaffolding (outlook_mail_extractor/config_migration.py) so legacy configs are upgraded in-place (v0 -> v2) with timestamped backups and jobs[].batch_flush_enabled default backfill.

Changed

  • Changed runtime path defaults for packaged/tool mode to user data directories (via platformdirs, with MAILSLIDE_DATA_DIR override), while preserving repo-root behavior for explicit source-mode runtime contexts.
  • Changed CLI default --config path resolution to use runtime context (runtime.paths.config_file) instead of hardcoded config/config.yaml.
  • Changed docs (README*, GUIDE*, task plans/checklists) to make PyPI/uv tool the primary distribution path and de-emphasize GitHub Releases as a release channel.
  • Changed About-tab Initialize Config feedback to use Textual notify(...) toast messages instead of inline status replacement, so users can reliably see setup results.
  • Changed Home-tab Preserve RE/FW toggle default to ON (while keeping the existing one-click toggle behavior) so first-run execution preserves reply/forward context by default.
  • Changed docs (README*, GUIDE*) to explicitly document Home Preserve RE/FW default-on behavior.
  • Changed README language filenames so English is now README.md and Traditional Chinese is README.zh-TW.md (previously README.en.md and README.md).
  • Changed LLM config save/load flow to keep api_key blank in config/llm-config.yaml, transparently loading from DPAPI secret storage when available and scrubbing legacy plaintext keys before backup writes.
  • Changed update notifications to use notify(...) when newer PyPI stable versions are detected, guiding users to upgrade with uv tool upgrade mailslide.
  • Changed load_config to auto-run config migration before validation and persist config_version back to config/config.yaml after successful migration.

Fixed

  • Fixed Home-tab run animation alignment/rendering so the ASCII block is right-aligned as a single block (without shifting the jobs table) and literal [@] characters render correctly instead of being interpreted as Rich markup.
  • Fixed plugin output-file discoverability in uv tool runtime by resolving relative output_dir / output_file values against the active config directory (not process CWD), preventing writes to unexpected locations.
  • Fixed plugin success observability for writable outputs by attaching output path details to summary_file / event_table success results.
  • Fixed first-run initialization UX by adding an explicit restart reminder toast after creating config files from About > Initialize Config.
  • Fixed plugin-config editor path-field clarity by converting relative output_dir / output_file values to absolute paths (based on active config directory) before editing/saving, so users can see and persist exact output locations.

[v0.3.5] - 2026-03-27

Changed

  • Updated both README language variants (README.md, README.en.md) with a clear AI-processing warning to avoid sending emails containing personal privacy data or business confidential information.
  • Updated both README language variants to explicitly recommend using local models when handling sensitive emails.
  • Added an explicit disclaimer in both README files stating that the project is provided as-is and project authors are not liable for data breaches or confidential information leakage caused by user inputs or model services.
  • Updated i18n default behavior to use en-US as the base language while auto-detecting Windows UI locale on first run; when locale resolves to Traditional Chinese, the app now defaults to zh-TW if no explicit/configured language is set.
  • Updated plugin config examples to English-first content across built-in plugin samples (add_category, move_to_folder, create_appointment, event_table, summary_file, write_file), including system prompts, profile descriptions, JSON example values, and _ui labels/messages.
  • Updated language-sensitive tests to accept bilingual validation messages where applicable and added coverage for first-run system-language fallback in TUI language initialization.

[v0.3.4] - 2026-03-27

Changed

  • Updated event_table Excel schema field name from outlook_link to outlook_open_command, and changed output behavior from clickable deep links/launcher files to plain PowerShell open commands for better compatibility in locked-down environments.
  • Updated event_table open-command generation to use PowerShell -EncodedCommand and robust Outlook COM search logic based on subject/received time traversal across stores/folders instead of relying on unstable EntryID/protocol handlers.
  • Updated default open-command matching window to subject + received time (within 1 day) and added optional sender matching control.

Added

  • Added open_command_match_sender plugin config (and _ui schema field) for event_table so users can optionally require sender matching when resolving mails from outlook_open_command.
  • Added EmailDTO.store_id and EmailDTO.internet_message_id extraction plumbing in core email extraction to support cross-store identification and future fallback matching strategies.
  • Added/updated event_table tests to cover encoded PowerShell command output and sender-match toggle behavior.

[v0.3.3] - 2026-03-27

Fixed

  • Fixed Home/Job execution logs so runtime messages now follow the selected UI language (ui_language) instead of staying hardcoded in Traditional Chinese.

Changed

  • Migrated job-run related log messages to i18n keys across runtime modules (job_execution, core, llm_dispatcher, logger) and added matching zh-TW/en-US locale entries.
  • Updated i18n test expectations in tests/test_i18n.py to align with current app branding/copy (app.title, app.subtitle).

Added

  • Added tests/test_job_log_i18n.py regression coverage to verify log message localization changes when switching between zh-TW and en-US.

[v0.3.2] - 2026-03-26

Changed

  • Renamed project/package distribution branding from outlook-mail-extractor to mailslide in package metadata, UI title strings, and README documents.
  • Replaced CLI command references with mailslide and removed the legacy outlook-extract script entry from packaging.
  • Updated About screen repository URL to https://github.com/linax777/mailslide.
  • Added Phase 1 import-path compatibility by introducing mailslide package re-exports while keeping outlook_mail_extractor available during migration.

Licensing

  • Declared package licensing metadata as GPL-3.0-or-later and updated user-facing license text to match the existing GPLv3 license file.

[v0.3.1] - 2026-03-26

Added

  • Added Phase 1 architecture split modules: outlook_mail_extractor/llm_dispatcher.py, outlook_mail_extractor/plugin_runner.py, and outlook_mail_extractor/move_policy.py to isolate LLM dispatch, plugin execution normalization/error wrapping, and move-target policy decisions.
  • Added plugin lifecycle hooks (begin_job / end_job) to support job-scoped buffering and flush workflows for writable plugins.
  • Added optional parser-level HTML artifact reuse (parse_email_html(..., use_cache=True)) so body extraction and table parsing can share one HTML parse per email.
  • Added metrics logging payloads for observability: per-mail summary (METRIC mail_summary) and per-job summaries (METRIC job_summary, METRIC job_execution) including elapsed time, LLM call counts, and plugin status distributions.
  • Added UI schema evaluator registry APIs (register_rule_evaluator, get_rule_evaluator, list_rule_evaluators) to support extensible rule wiring without editing the core evaluator map.
  • Added optional dynamic plugin discovery support via top-level config plugin_modules (runtime import for custom plugin modules).
  • Added new tests for Phase 1/2 behavior, including tests/test_llm_dispatcher.py, tests/test_plugin_runner.py, tests/test_move_policy.py, parser HTML parse bundle coverage, plugin batch-flush workflows, and runtime rule-registry registration.

Changed

  • Refactored EmailProcessor orchestration to call shared dispatch/execution/policy helpers instead of duplicating shared/per-plugin/no-LLM branches, while preserving existing external behavior.
  • Updated EmailProcessor.extract_email_data to parse HTML once and reuse parsed artifacts for cleaned body and table extraction.
  • Updated event_table and summary_file output strategy from immediate per-mail writes to default job-level batch flush (configurable per job with batch_flush_enabled, default true, with fallback switch-off support).
  • Updated changelog/version consistency by aligning package __version__ with pyproject.toml and documenting current config/runtime knobs in samples and README.
  • Updated workspace hygiene by ignoring *.bak files to reduce backup-noise in local working trees.

[v0.3.0] - 2026-03-24

Added

  • Added English documentation file README.en.md with installation, TUI/CLI usage, configuration examples, plugin overview, localization notes, and local llama.cpp setup.
  • Added cross-language navigation links at the top of both README files (README.md <-> README.en.md) so users can switch languages quickly.
  • Added usage-screen tests in tests/test_usage_screen.py to verify language-based README resolution and fallback behavior.
  • Added key-based i18n runtime (outlook_mail_extractor/i18n.py) with gettext catalog loading and YAML fallback dictionaries for zh-TW and en-US.
  • Added CLI language override --lang (zh-TW/en-US) and config-level ui_language support in config/config.yaml(.sample).
  • Added Language modal in TUI (hotkey L) with radio-button selection, persisted language updates to config/config.yaml, and immediate UI recompose on language switch.
  • Added Babel i18n tooling (babel.cfg, locale pot/po files, scripts/i18n.ps1) and README instructions for extract/init/update/compile workflows.
  • Added i18n/config tests (tests/test_i18n.py, tests/test_config_ui_language.py, tests/test_app_language.py) and extended schema tests for label_key/message_key support.

Changed

  • Migrated major TUI/CLI surfaces to translation keys, including app shell/tab labels, Home/About/Schedule/Usage pages, Configuration tabs, Main Config workflows, and key modal/error notifications.
  • Updated _ui schema handling to support localized key fields (label_key, message_key) with backward-compatible fallback to raw label/message text.
  • Updated Plugin/Main config editor rendering and rule evaluation paths to resolve translated schema labels/messages at runtime.
  • Updated packaging metadata to include locale resources and Babel dev dependency.
  • Updated Guide tab content loading (outlook_mail_extractor/screens/usage.py) to read README by active app language: en-US prefers README.en.md with fallback to README.md, while zh-TW continues using README.md.

[v0.2.8] - 2026-03-24

Added

  • Added per-job manual_review_destination routing support so emails with no LLM action outcome are moved to a dedicated manual-review folder.
  • Added high-risk core tests to cover move routing for LLM success -> destination, non-action -> manual_review_destination, and plugin failure -> manual_review_destination.
  • Added a new 修改 Job action in Main Configuration _ui schema so users can update existing jobs from the form-driven workflow instead of editing YAML manually.
  • Added modal/form test coverage for job edit helpers and add/edit payload handling, including manual_review_destination field mapping.

Changed

  • Updated orchestrator move logic to route mails based on LLM plugin result statuses: SUCCESS goes to destination, while SKIPPED/FAILED/RETRIABLE_FAILED can be routed to manual_review_destination.
  • Updated config sample/UI schema and README job fields to document and expose manual_review_destination in TUI and YAML workflows.
  • Updated Main Configuration tab to support editing the selected job through a dedicated modal, reusing the structured job form fields (name/account/source/destination/manual_review_destination/limit/plugins) and reducing direct YAML edits.

[v0.2.7] - 2026-03-24

Changed

  • Updated event_table output from CSV to Excel (.xlsx) and changed default output path to output/events.xlsx.
  • Updated event_table fixed schema to include Outlook message identifiers and deep-link fields: email_entry_id and outlook_link.
  • Updated config/plugins/event_table.yaml(.sample) UI schema and validation rule from .csv to .xlsx output checks.
  • Updated README plugin documentation for the new Excel-based event table workflow and Outlook classic open-link behavior.

Added

  • Added EmailDTO.entry_id and wired message extraction to include Outlook EntryID in core processing.
  • Added Excel hyperlink writing in event_table so each row includes a clickable outlook:<EntryID> link (Open in Outlook) for opening the matching mail in Outlook classic.
  • Added openpyxl runtime dependency and lockfile update for Excel read/write support.
  • Added/updated tests covering Excel output headers/rows, hyperlink generation, append behavior, timezone datetime parsing, and UI schema rule support for output_file_xlsx.

[v0.2.6] - 2026-03-23

Changed

  • Changed LLM plugin orchestration default to per_plugin, so each LLM-enabled plugin now gets its own llm_client.chat(...) call and response parsing path, avoiding cross-plugin action collisions when multiple plugins are enabled in one job.
  • Added backward-compatible legacy mode share_deprecated that restores shared single-response behavior for migration scenarios, with explicit runtime warnings that this mode can cause action_mismatch/skipped outcomes across mixed-action plugins.
  • Added llm_mode support at both config scopes (config.llm_mode default + job.llm_mode override), with runtime resolution and alias mapping from legacy values (shared, shared_legacy) to share_deprecated.
  • Updated job execution wiring to pass global llm_mode defaults into processor dispatch while allowing per-job overrides without changing existing plugin config loading behavior.
  • Updated config/config.yaml.sample and README documentation to describe llm_mode, recommended default usage (per_plugin), and deprecation guidance for shared-response mode.

Added

  • Added config validation rules for llm_mode values in both top-level config and per-job settings, including backward-compatible acceptance of legacy alias values.
  • Added high-risk orchestration tests covering per-plugin independent LLM calls, share_deprecated single-call behavior, and alias compatibility (shared -> share_deprecated) in tests/test_core_high_risk.py.

[v0.2.4] - 2026-03-23

[v0.2.5] - 2026-03-23

Added

  • Added plugin prompt profile support so a single plugin config can define multiple prompt variants (prompt_profiles) with default_prompt_profile, and each job can select a profile via plugin_prompt_profiles.
  • Added prompt-profile handling tests in tests/test_core_high_risk.py covering profile precedence, default fallback, missing-profile fallback warnings, shorthand profile values, and config immutability.
  • Added yaml field-type tests for plugin editor payload collection, including valid parse, empty input handling, and invalid YAML errors.

Changed

  • Updated plugin runtime prompt resolution to apply profile-specific prompt overrides before plugin instantiation, while preserving legacy system_prompt behavior as final fallback.
  • Updated config/config.yaml.sample and config/plugins/add_category.yaml.sample to document and demonstrate prompt-profile configuration.
  • Extended prompt-profile sample/UI schema coverage to all built-in LLM plugins (move_to_folder, create_appointment, event_table, summary_file), so Plugin Configuration modal now shows the same OptionList-based profile editor for each of them.
  • Updated Plugin Configuration modal UX for prompt_profiles editing: replaced raw YAML-only workflow with an OptionList-based profile switcher and per-profile detail editor (version, description, system_prompt).
  • Added profile management controls in Plugin Configuration modal (+ 新增 / - 刪除) to create and remove prompt profiles interactively while enforcing at least one remaining profile.
  • Updated config/config.yaml.sample to show plugin_prompt_profiles under each job (per-job scope) and removed the misleading top-level example.
  • Updated README with a dedicated "同 Plugin 多 Prompt（Prompt Profiles）" section describing profile structure, job-level selection, and resolution order.

[v0.2.4] - 2026-03-23

Fixed

  • Fixed Plugin Configuration tab file discovery so backup files (*.yaml.bak) are no longer treated as editable plugin entries, preventing phantom *.yaml rows that failed with "file not found" when opened.

Changed

  • Added a new Plugin tab action button 🧹 清理備份 to remove stale plugin backup files (config/plugins/*.yaml.bak) and refresh the plugin list in place.
  • Extended plugin-config tests to cover both backup-file filtering in the plugin list and backup cleanup behavior.

[v0.2.3] - 2026-03-21

Fixed

  • Fixed reply/forward body cleanup so metadata header lines are consistently removed from multi-layer threads (From/Sent/Date/To/Cc/Subject and 寄件者/已傳送/寄件日期/收件者/收件人/副本/主旨) even when preserving RE/FW thread content.
  • Fixed reply-thread trimming behavior to keep meaningful older body text instead of dropping everything after the first separator block in common reply chains.

Changed

  • Expanded parser header pattern matching to cover colon-spacing and mixed Chinese/English variants such as 副本(CC) and Date: for more stable LLM input cleanup.
  • Added parser debug observability by logging pre/post cleanup body lengths during email extraction to help verify noise-reduction impact.
  • Added/updated parser tests for multi-layer metadata stripping and date-header variants; current parser suite now validates this cleanup path with uv run pytest -q tests/test_parser.py.

[v0.2.2] - 2026-03-20

Highlights

  • Refactored the large monolithic screens.py into a modular outlook_mail_extractor/screens/ package while keeping from outlook_mail_extractor.screens import ... imports backward-compatible.
  • Improved maintainability of schema-driven configuration UI by extracting shared YAML write/backup and validation-message helpers used across main/LLM/plugin config tabs.
  • Simplified PluginConfigEditorModal internals by splitting dynamic form rendering and payload collection into smaller focused methods without changing runtime behavior.

Changed

  • Reorganized UI screens into dedicated modules (about, home, schedule, usage, config tabs, and modals) and removed the legacy single-file outlook_mail_extractor/screens.py implementation.
  • Added config-tab helper modules for consistent YAML serialization/atomic write+backup behavior and reusable rule failure preview formatting.
  • Kept existing TUI workflows and tests stable through refactor with no user-facing feature regression.

[v0.2.1] - 2026-03-19

Highlights

  • Added a schema-driven configuration editing workflow in the TUI so config samples can define action buttons, field metadata, and validation rules while runtime configs remain user-editable YAML.
  • Improved job authoring UX by replacing direct empty-row insertion with a modal add-job form that validates required fields before updating config content.
  • Added the new summary_file plugin to support LLM-generated summary rows appended to a CSV output file.

Changed

  • Extended config/*.yaml.sample and config/plugins/*.yaml.sample with _ui metadata blocks for future form generation and centralized validation behavior.
  • Updated the Configuration tab layout to prioritize the jobs list and editable config area, with schema action buttons for validate/save/add/remove/reset.
  • Switched add-job plugin selection to Textual SelectionList driven by plugin options declared in config schema.
  • Updated the Plugin Configuration tab to support schema-driven modal editing (str/int/bool/select/textarea/path/list[str]/multiselect), with field-level validation and _ui.validation_rules execution before save.
  • Added plugin-config save safeguards: remove reserved metadata keys before write, and auto-create <plugin>.yaml.bak when overwriting an existing file.
  • Refined response_json_format editing UX: keep action/start/end locked as fixed template keys while allowing other JSON example fields to be edited safely as structured inputs.
  • Removed the unused LLM provider runtime setting from config/UI samples and status display, since execution already uses OpenAI-compatible api_base directly.
  • Updated README with _ui metadata guidance and summary_file plugin usage examples.
  • Hardened Main Config save flow in the TUI to validate editor payloads before write (_ui.validation_rules plus runtime validate_config), preventing invalid YAML objects from overwriting config/config.yaml.
  • Added Main Config backup+atomic write behavior (config.yaml.bak + temp-file replace), aligning safety guarantees with existing LLM/plugin editor writes.
  • Improved Main Config remove-job UX to delete the selected row when available (with fallback to last item) and updated the notification to show both removed row number and job name.

Added

  • Added outlook_mail_extractor/ui_schema.py helpers for loading schema metadata, flattening field definitions, evaluating validation rules, generating list-item defaults, and stripping reserved keys.
  • Added load_plugin_ui_schema and Plugin Configuration TUI wiring to load plugin _ui schemas from config/plugins/*.yaml.sample with read-only fallback when schema is missing.
  • Added config/plugins/summary_file.yaml.sample, outlook_mail_extractor/plugins/summary_file.py, and tests/test_summary_file_plugin.py for the summary CSV plugin workflow.
  • Added tests/test_ui_schema.py coverage for schema field flattening, rule evaluation, default list-item generation, reserved-key stripping, and plugin schema loading.
  • Added tests/test_plugin_config_editor.py coverage for modal payload mapping/validation and plugin config file backup-write behavior.
  • Added tests/test_main_config_editor.py coverage for Main Config editor payload validation and backup/atomic write behavior.

[v0.2.0] - 2026-03-19

Highlights

  • Completed architecture improvement milestones PR-1 through PR-7, including service extraction, typed error layering, structured plugin results, capability-based plugin dispatch, DTO/action-port boundaries, and runtime-context dependency injection.
  • Unified CLI/TUI behavior through shared execution and preflight services, reducing duplicated UI business logic and improving consistency.
  • Improved testability and maintainability by replacing hardcoded paths/global state with injectable runtime dependencies and by expanding automated test coverage.

Fixed

  • Fixed event_table CSV schema handling by making column definitions code-fixed and ignoring user-provided fields overrides, with a warning log for deprecated config keys.
  • Fixed event_table datetime parsing to accept ISO8601 timestamps with timezone offsets (for example +08:00) and UTC Z suffixes.
  • Fixed create_appointment datetime parsing to accept ISO8601 timezone timestamps and normalize them to Outlook-COM-safe naive datetimes before appointment creation.

Changed

  • Introduced JobExecutionService to own config-driven job orchestration, and refactored core.process_config_file into a thin compatibility wrapper while aligning CLI/TUI execution paths to the shared service.
  • Added plugin capability-based dispatch (requires_llm, can_skip_by_response, moves_message) so orchestrator behavior is declared by plugin capabilities instead of hardcoded plugin names.
  • Updated built-in plugins to declare capabilities and moved appointment skip logic (create=false) into the create_appointment plugin via a response-based skip hook.
  • Added mypy to the dev dependency group, installed typing stubs (types-pyyaml, types-pywin32), and introduced project mypy config (including pycron missing-stub override) so static typing checks can run in CI/local workflows.
  • Applied targeted typing improvements in core/services/tests to make uv run mypy . pass without changing runtime behavior.
  • Added plugin tests covering timezone datetime handling in both event_table CSV export and create_appointment calendar creation flows.
  • Removed the legacy outlook_worker.py compatibility script because the project now uses outlook_mail_extractor.__main__ as the single CLI entry point.
  • Updated .gitignore and git tracking so local workflow notes (AGENTS.md, DEVELOPER.md, tasks/) stay untracked.
  • Introduced application-level error layering (AppError, DomainError, InfrastructureError, UserVisibleError) and aligned core processing error handling to preserve typed failures while wrapping unknown exceptions as infrastructure-level errors.
  • Introduced structured plugin execution results via PluginExecutionResult and PluginExecutionStatus (success, skipped, failed, retriable_failed) to improve plugin observability and diagnostics.
  • Updated EmailProcessor orchestration to normalize both legacy bool plugin returns and new structured results, preserving backward compatibility while standardizing plugin logs and result metadata.
  • Migrated built-in plugins (create_appointment, move_to_folder, add_category, event_table, write_file) to return structured execution outcomes with explicit status codes/messages.
  • Added a shared PreflightCheckService for validating enabled jobs against Outlook account/source availability, and refactored TUI pre-run + About checks to use this common service.
  • Added CLI preflight validation before job execution (with --skip-preflight escape hatch) so command-line runs now catch account/source config issues early.
  • Added unit tests for preflight validation behavior (tests/test_preflight_service.py) and cleaned related UI lint warnings while wiring the new service.
  • Introduced EmailDTO and MailActionPort to separate domain email data from Outlook COM side effects, and added OutlookMailActionAdapter as the infrastructure implementation.
  • Refactored EmailProcessor and built-in plugins to use the action port boundary instead of hidden _message/_account fields, so plugin behavior can be tested with fake ports without COM dependencies.
  • Updated plugin/core tests to validate the new DTO + action-port flow and preserve existing processing behavior.
  • Introduced runtime dependency wiring via RuntimeContext/RuntimePaths and updated CLI/TUI flows to consume injected paths, logger manager, and client factory instead of hardcoded module-level paths.
  • Refactored logger state into an injectable LogSessionManager while keeping LoggerManager as a backward-compatible static facade.
  • Extended core.process_config_file and JobExecutionService dependency injection points (runtime paths + logger + factories/loaders), reducing monkeypatch needs in tests.

[2026-03-18]

Fixed

  • Fixed OpenAI-compatible endpoint joining in the LLM client by using a relative chat/completions path, so configured api_base paths (for example /v1) are no longer dropped.
  • Fixed LLM API base handling by normalizing api_base (trim + single trailing slash), preventing malformed URL joins across local providers such as LM Studio.
  • Fixed LLM HTTP error reporting to include server-side error details and request URL context, making 400 and upstream timeout (504) diagnosis actionable in UI/CLI logs.
  • Fixed Home-tab execution controls by adding a dedicated stop button that cancels the running worker and restores run/stop button states cleanly after cancellation or completion.
  • Fixed job execution flow so destination moving still works when no LLM client is available and no-LLM plugins (such as write_file) still execute instead of exiting early.
  • Fixed duplicate move behavior by skipping final destination moving when move_to_folder has already moved the message.
  • Fixed config-relative loading by resolving llm-config.yaml and plugins/ from the provided --config directory first, with fallback to default config/ paths.
  • Fixed parser reply-history behavior so preserve_reply_thread=True now keeps reply metadata lines instead of removing them unconditionally.
  • Fixed parser style handling for tags with non-string style attributes to avoid attribute access/type-check issues during HTML cleanup.

Changed

  • Added pytest to the dev dependency group and lockfile so uv run pytest works without temporary dependency flags.
  • Added a new event_table plugin that reuses appointment-style LLM extraction and appends event rows to a single CSV table (output/events.csv) instead of creating Outlook calendar items.
  • Added config/plugins/event_table.yaml.sample and README documentation for configuring CSV output path and column order.
  • Updated the About screen version label to 0.1.7.
  • Updated .gitignore to keep user-local config/*.yaml and config/plugins/*.yaml ignored while allowing tests/ to be tracked in git.

[2026-03-17]

Fixed

  • Fixed the outlook-extract console entry point so the CLI runs correctly instead of returning an un-awaited coroutine.
  • Fixed --no-move so CLI runs can extract mail data without moving processed messages.
  • Fixed cleanup in the processing pipeline to avoid masking the original exception when Outlook or LLM setup fails early.
  • Fixed calendar creation to resolve the default Calendar folder from the configured Outlook account instead of the global default mailbox.
  • Fixed the scheduler lifecycle in the Textual UI to stop duplicate interval timers from accumulating and to shut them down cleanly on unmount.
  • Fixed the scheduler shutdown crash caused by _schedule_timer being initialized on the wrong screen class.
  • Fixed job message limiting so limit now counts actual mail items only, while folders with no mail items complete normally without raising errors.
  • Fixed Outlook account resolution to use strict store-name matching, so invalid account names are detected reliably instead of being accepted by ambiguous COM lookup behavior.
  • Fixed the no-LLM plugin execution path (e.g., write_file) so processed messages are still moved to the configured destination folder.
  • Fixed an UnboundLocalError in the no-LLM path by initializing error_msg before early return branches.

Changed

  • Improved email body extraction to prefer cleaned HTML content when available, preserve paragraph structure, and trim quoted reply history before sending content to the LLM pipeline.
  • Updated parser body cleanup to remove reply/forward metadata header lines from content (From/Sent/To/Cc/Subject and 寄件者/已傳送/收件者/副本/主旨) to reduce body noise and length.
  • Improved parser cleanup with high-confidence signature and newsletter footer stripping to reduce noise in extracted email bodies.
  • Improved forwarded-message parsing so short FW:/Fwd:/轉寄 comments keep the forwarded body instead of truncating everything after the first forwarded header block.
  • Added parser coverage for forwarded-message extraction and standard reply-thread trimming behavior.
  • Improved first-run onboarding in the TUI so missing config/config.yaml now redirects users to the About tab, shows initialization guidance, and disables job execution until setup is completed.
  • Improved validation in the About tab so system checks now verify enabled jobs' account and source settings before execution.
  • Improved job execution so configured destination folders are created automatically when missing, which helps first-time runs on new hosts.
  • Changed parser defaults to preserve RE/FW thread content unless explicitly disabled via preserve_reply_thread=False.
  • Added a Home-tab toggle control for preserving RE/FW content during parsing and wired it through to the processing pipeline.
  • Replaced the Home-tab switch widget with a visible ON/OFF toggle button to avoid terminal layout visibility issues.
  • Added configurable body_max_length support in config.yaml (global default plus per-job override) so body truncation length can be tuned by environment.
  • Updated config validation to enforce positive-integer body_max_length values.
  • Updated config/config.yaml.sample with documented body_max_length examples.
  • Updated .gitignore to ignore local tests/ and tasks/ directories.