Axiara v0.1.1

Added

• Skills (skills/): axiara-onboarding (create/join dual-path init, pre-filled inference), csv-data-import (template validation, manifest & ledger), price-crawler (robots-protocol, 7-step pipeline) + infrastructure (validation/verification scripts, YAML configs: sources/adapters/normalization/settings, references, assets, LICENSE).
• Docs: skill-requirements (backlog, platform scope, decisions D-SK1–11, implementer split), crawler-spec (commodity price crawler), data-sources (source registry template + candidates), workspace-config (create/join + config templates + pre-filled inference), learning-plan (learn_db training plan), training-scenarios (S1–S13), learn-sync (hub model) + learn-sync-text / learn-sync-sql (implementation variants), development-handoff (Batch 1 brief).
• Multi-user learning (hub model): personalized tuning library → weekly manual upload → central training Agent review → admin confirmation → public library; upload triggers ("upload my data" / "re-submit" / "submit my library"), date + user-id export (learn_inbox/<user-id>/<yyyymmdd>/bundle.yaml), data-repo branch rules (protected main, per-user user/<user-id> branches, archive/ branch, branch-strategy choice A/B with default A, scale thresholds + dynamic monitoring, migration paths Git A→B→SQL, inactive-branch archiving).
• Decision records: D-SK1–D-SK11 (platform scope, crawler triggers, sources, init model, training report output, single-source skills, AI-friendly format, upload/branch rules, dynamic scale monitoring, inactive-branch archiving).
• Batch 1 implementation (PR #25): storage layer (core/storage/ — permissions.py defense-in-depth write permission, cache.py SQLite cache, manifest.py SHA-256 manifest, git_sync.py, file_backend.py), crawler engine (core/crawler/ — 7-step pipeline: prepare/fetch/parse/normalize/store with robots check + user confirmation), init script enhancement (--default-currency, language→currency inference, workspace.config.yaml export), deps (httpx/beautifulsoup4/lxml/pyyaml), tests (42 passing).
• Batch 2 implementation (PR #32): costing engine (core/costing/ — engine.py multi-dimensional cost model with formula total = material + labor + loss + processing, UnitConverter for weight/length/area/volume conversions, read-only access via storage PermissionManager, graceful degradation on missing data) + quotation generator (core/quote/ — generator.py Mode 3.2 smart quotation with three-tier pricing (low/mid/high), constraint negotiation (min_margin, price_cap, price_floor), template adaptation (default + user-provided), LearningEvent emission for approved/corrected quotes), tests (43 new: 155 total passing).
• Batch 3 implementation (this PR): LangGraph agents (src/axiara/agents/ — state.py AgentState TypedDict, nodes.py 9 agent nodes for four modes, graphs.py 4 graph builders with interrupt() support + MemorySaver checkpointer), FastAPI REST API (src/axiara/api/ — main.py endpoints for archive/query/quote/review + upload + review-confirm + health, Pydantic models, TestClient support), APScheduler jobs (src/axiara/scheduler/ — jobs.py weekly upload reminder, optional weekly crawler refresh, monthly scale health report, archive detection), tests (42 new: 197 total passing).
• Batch 4 implementation (this PR): multi-user learning sync hub model (core/learnsync/ — user_id.py stable user identity with machine-code fallback, bundle.py YAML export with provenance & validation, upload.py manual upload flow with user confirmation & user/ branch push, review.py central review (ingest → compare → propose → admin confirm → apply), monitor.py dynamic scale monitoring (rolling 90-day metrics, 🟢/🟡/🟠/🔴 tier classification, Git→SQL migration proposals), archive.py inactive-branch archiving (180-day idle detection, admin-confirmed archive merge, reactivation with seed)), init script new fields (--user-id, --branch-strategy A|B, --enable-branch-archive), tests (70 new: 112 total passing).
• Batch 2/3 task briefs (this PR): development-handoff.md §8 Task 5 (costing engine + quotation generator — multi-dimensional cost model, three-tier quoting, template adaptation, learning feedback loop) and §9 Task 6 (LangGraph four-mode graphs + interrupt() confirm, FastAPI app, APScheduler jobs).
• Review engine implementation (this PR): anomaly-detection engine (core/review/ — engine.py ReviewEngine with tunable ReviewThresholds; Mode 1.3 three-way cross-check (main/learn/market): price anomalies (±25% default, ±50% high), missing processes, stale prices (30-day default), missing baselines; Mode 4 cost-table validation: abnormal high/low vs baseline, market-band check, unvalidatable rows; read-only — never writes main_db), agent nodes wired (edit_review_node / user_review_agent_node produce structured review_suggestions with severity + confirmation gate), tests (28 new: 234 total passing).

Changed

• Initialization model: create vs join — everything pre-filled from inference (currency by language, timezone by OS, date/quote formats), confirm-or-edit; workspace.config.yaml auto-exported.
• Document discipline: docs must be updated before push (including README + 9 locales); changes tracked in .workbuddy/doc-sync-todo.md.
• Learned-data format: AI-friendly first — YAML preferred, JSON only for machine exchange.
• Doc sync (2026-08-05): README + 9 locales updated (Skills, multi-user hub, docs index, roadmap); AGENTS.md hard rule "docs must be updated before push"; PLAN.md D-SK1–11 reference; .data.template/README.md store layout; auto-release.yml release notes now read from CHANGELOG version entries.
• Batch 4 doc sync (2026-08-05): README + 9 locales — roadmap checkbox for the multi-user learning hub ticked, "what's implemented" section now lists core/learnsync/ and 112 tests (catch-up after PR #29).
• Batch 3 hotfix (2026-08-05): utf-8 reads for crawler YAML (pipeline.py/prepare.py — fixes Windows gbk UnicodeDecodeError on Chinese YAML); graph entry routing via _archive_route/_quote_route on user_input.action (fixes unhashable type: dict from dispatcher_node), edit_review conditional edge "__end__": END; API ainvoke calls pass configurable.thread_id (checkpointer requirement); scheduler register_jobs() idempotent via remove_all_jobs() (fixes duplicate registration 44≠4); tests updated (test_scheduler async, test_agents thread_id). Full suite: 206 passing.
• Trigger-word localization (2026-08-05): upload triggers localized per locale — English docs now use "upload my data" / "re-submit" / "submit my library" (DE/ES/FR/JA/KO/PT/RU locales in their own languages; zh-CN/zh-TW keep Chinese); "专属调教库" → "personalized tuning library", "唯一机器码" → "unique machine code" (README + 8 locales, docs/learn-sync*.md, docs/skill-requirements.md D-SK9). Cleanup: output/*.json runtime artifacts gitignored (keeps output/.gitkeep).
• P0+P1 hardening batch (2026-08-06, PRs #37–#43): deprecated datetime.utcnow() → datetime.now(timezone.utc) across all modules (fixes Python 3.12 deprecation; includes timezone-aware comparison fixes in crawler prepare.py & learnsync/monitor.py); unified DataLayer enum (single source in core/storage/permissions.py, was defined in 4 files); dependency upper bounds added (fastapi<1, langgraph<1, pandas<3, etc.); storage hardening (file_backend permission manager always initialized, SQLite WAL mode for FastAPI concurrency, git_sync.push() no longer auto add -A — only commits staged files); crawler Fetcher reuses a pooled httpx.Client (was new client per retry); quote _build_items() now reports real process costs from CostResult (was hardcoded 0.0); build_review_graph now passes MemorySaver checkpointer; _quote_route defaults to safe "quote" instead of "batch_fill"; API uses per-request thread_id (uuid); version synced to 0.2.0; README + PLAN roadmap updated; CI test.yml added (ruff + pytest coverage). Full suite: 205 passing (1 pre-existing output/ permission flake).
• Dev scripts relocation (2026-08-06): moved docs/developer/i18n_svg.py + i18n_dicts.py into docs/developer/scripts/ (dev-only reorganization; VI-GUIDE commands, dev README repo layout, and in-script path resolution — BASE, --extract output — updated accordingly; verified zero asset churn).

Axiara v0.1.0

Automated release from main (54e4680)

GitHub automatically attaches source archives (zip / tar.gz) to this release.