Feat/restart no nanoclaw#2476
Open
andrefogelman wants to merge 224 commits into
Open
Conversation
…RAG, flow engine - Add src/supabase-client.ts: bridge with supabase-js + WebSocket polyfill - Add src/anf/: ANF agent modules (activity-log, agent-context, agent-registry, agent-runner, chat-handler, embeddings, flow-engine, health, memory, realtime-listener, router, scheduler, task-executor, whatsapp) - Add src/anf/anf-config.ts: ANF config (Anthropic base URL, port, admin WA) - Add src/anf/start.ts: ANF entry point with scheduled tasks for all 4 agents - Add agents/: 4 domain agents (suprimentos, financeiro, engenharia, orquestrador) - Fix all imports: db.js → supabase-client.js, channels/whatsapp.js → whatsapp.js - Update tsconfig.json: rootDir=. and include agents/**/* - Add .env.example with all required env vars Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- startAnfModules() chamado no main() após init - NanoClaw roda em ANF-only mode quando sem channels - Subsystems NanoClaw (scheduler, IPC, message loop) só iniciam com channels - Módulos ANF (realtime, task executor, flow engine, chat, crons) rodam independente Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Rename package to orcabot v0.1.0 - Replace better-sqlite3 with @supabase/supabase-js, add pino, remove @onecli-sh/sdk - Create OrcaBot config with Supabase/LLM proxy/API env vars - Add OneCLI stub to keep existing code compiling - Add @ts-nocheck to db.ts and db-migration.test.ts (pending Task 4 rewrite) - Create directory structure: agents/, container/skills/, supabase/migrations/, tests/, frontend/ - Add .env.example with all env vars documented - Replace CLAUDE.md with OrcaBot-specific instructions Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
7 migration files for OrcaBot schema on shared Supabase (baebsednxclzqukzxkbg): - 001: ob_organizations, ob_org_members (auth & tenancy with RLS) - 002: ob_projects, ob_project_files (RLS via org membership) - 003: ob_pdf_jobs, ob_pdf_pages (PDF pipeline with RLS) - 004: ob_quantitativos, ob_orcamento_items (budget with RLS) - 005: ob_sinapi_composicoes (public read-only, pg_trgm fuzzy search), ob_cotacoes_mercado - 006: ob_agent_activity_log, ob_agent_conversations (RLS) - 007: ob_nc_* NanoClaw runtime tables (no RLS, service_role only) All tables use ob_ prefix to avoid conflicts with anfconstrucoes tables. Includes src/migrate.ts runner and "migrate" npm script. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two clients (supabaseAdmin for server-side, supabaseAnon for RLS), supabaseWithAuth(jwt) helper, legacy aliases for backward compat, and test suite. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Replace all better-sqlite3 usage with supabaseAdmin client - All db functions now async (return Promises) - Update all callers (index.ts, ipc.ts, task-scheduler.ts) to await - Update all tests to use async/await patterns - Remove db-migration.test.ts (SQLite migrations no longer apply) - Remove SQLite from setup/ files (environment, verify, groups, register) - Add .env loading to vitest.config.ts for test env vars - Zero SQLite references remaining in src/ and setup/ Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add normalizeTimestamp() to handle Postgres timestamptz +00:00 format - Fix test data with invalid dates (June 31+, Feb 30) for Postgres strict mode - Make IpcDeps.registerGroup async to properly await Supabase writes - Add fileParallelism: false to vitest config (shared remote DB) - Rename migrations to timestamp format for Supabase CLI compatibility - Add supabase config.toml and link to project Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Proxy listens on configurable port (default 8200) and forwards container LLM requests to Anthropic with the correct auth headers. Supports apikey mode (production) and max mode (dev, zero cost). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
HTTP server channel that bridges the OrcaBot frontend to NanoClaw. Endpoints: POST /api/message, POST /api/job, GET /api/status/:id, GET /api/health. Auth via Bearer token (API_SECRET). Self-registers with channel registry. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Dockerfile for main OrcaBot gateway, docker-compose with ports 8200/8300, agent container image with PaddleOCR/tesseract, and rsync deploy script. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Health check verifies Supabase, LLM proxy, and API channel connectivity. Integration tests validate ob_nc_chats CRUD and existence of ob_organizations, ob_projects, and ob_sinapi_composicoes tables. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Container skill with 5-stage pipeline: ingestion, extraction, classification, interpretation, and structured output. Includes pdfjs-dist text extraction, PaddleOCR fallback, Claude Vision interpretation, Zod-validated output schemas, and confidence scoring. All 24 unit tests passing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Shared infrastructure: types.ts + supabase-helpers.ts for all agents - Orcamentista (orchestrator): 8 tools (PDF results, quantitativos, delegation, SINAPI search, orcamento items, subtotals, review flags, project context) - Estrutural (specialist): 4 tools (concreto, aco, forma, escavacao) - Hidraulico (specialist): 3 tools (tubulacoes, conexoes, equipamentos) - Eletricista (specialist): 4 tools (pontos, circuitos, eletrodutos, materiais) - Agent registry with all 4 agents - Delegation engine with Supabase-based IPC (poller + task processor) - SINAPI import script (Python + openpyxl) - SINAPI lookup container skill - Migration for ob_delegation_tasks table - 37 tests across 9 test files, all passing Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…wer, chat, export) Complete web frontend with: - Auth (Supabase login/session), protected routes - Dashboard with project list, status cards, new project wizard - Budget spreadsheet with inline editing, 3-level hierarchy, ANF format - PDF upload (drag-and-drop), viewer (react-pdf), review panel - Agent chat sidebar with Realtime subscriptions - Quantitativos, Premissas, Curva ABC tabs - Excel export (ExcelJS) matching ANF format - Supabase Realtime for live updates on budget, PDFs, chat - shadcn/ui components, TailwindCSS, TypeScript strict Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Database tables use ob_ prefix (ob_projects, ob_orcamento_items, etc.) but frontend .from() calls and types referenced them without prefix. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add PDF job poller that processes pending jobs sequentially - Fix table name mismatches (ob_ prefix) in pdf-pipeline supabase client - Fix storage bucket creation and RLS policies for project-pdfs - Fix auth.users permission error in storage avatar policies - Implement delegation engine with real agent execution via Anthropic SDK - Add OrcaBot agent runner for specialist agents (estrutural, hidraulico, eletricista) - Implement Excel export skill with 3-sheet XLSX generation - Add export_budget tool to orcamentista agent - Create process-project Supabase Edge Function (serverless pipeline) - Add "Iniciar Orçamento" button to ProjectPage header - Wire job poller and delegation engine into main index.ts - Add retry with backoff for Claude API rate limits - Use King proxy (100.91.255.19:8100) for LLM calls Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…mpts - Add PdfProcessPanel: chat-like interface for processing individual PDFs - User selects a PDF, types what to extract, AI processes with that prompt - Smart prompt suggestions based on filename (demolir, hidraulica, eletrica, etc.) - Shows extracted results inline with confidence badges - Replace ReviewPanel with PdfProcessPanel in PdfsTab - Create process-single-pdf Edge Function for per-file processing - Retry with backoff for rate limits Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…Funnel - Add item selection table in PdfProcessPanel after extraction - Checkbox per item, select all/none controls - "Salvar na Planilha" button inserts selected items into ob_quantitativos - No automatic SINAPI lookup — user fills costs manually - Fix edge functions: use pdf-parse instead of pdfjs-dist (Deno compat) - Add node:buffer import for Deno edge functions - Route LLM calls through King Tailscale Funnel (zero API cost) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Create ob_processing_runs table to store every AI extraction attempt - Each run saves: prompt, items, summary, needs_review, raw_response - PdfProcessPanel shows full history of runs per file (expandable cards) - Each run card shows items table with select/save to planilha - Drastically improved system prompt: forces numeric quantities, includes calculation rules for walls/floors/pipes, never returns empty items - Increased max_tokens to 8192 for longer extractions - Warns when PDF has very little extractable text (scanned image) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Paginated table with unit-match highlight (✓ badge), click to select, double-click equivalent. Uses existing useSinapiSearch hook. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Mirror of SINAPI tab without pagination (1272 total, hook limits 100). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds BookOpen button per row that opens a shared PriceSourceDialog hosted by BudgetTable. On apply, pushes an 'update' undo entry with the previous cost snapshot so Ctrl+Z reverts cleanly. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Was invisible with group-hover:visible, but design spec calls for always-visible affordance (inline UI pattern). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Keep safeJsonParse (upstream), taskContent param name (upstream), and console.error logging (upstream) over stashed local changes. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Previously custo_total was set to just material + mao_obra without multiplying by quantidade. Also fix TCPO search to match on descricao and search_term (accent-insensitive). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Previously deleting a level-2 item would leave its level-3 children orphaned. Now all descendants are deleted regardless of parent level. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When editing an etapa/item EAP code (e.g. 08 → 07), all children are automatically renumbered (08.01 → 07.01, etc.) using the existing renumber_eap_items RPC. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The React Query cache wasn't refreshed after the renumber RPC, so the user had to manually reload to see updated EAP codes. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- BudgetCell: single-click focus, type to edit, arrow/tab/enter navigation - BudgetRow: pass focus state and navigation callbacks to each cell - BudgetTable: manage focused cell state, flat row list, keyboard handler Click selects cell, typing starts editing, arrows navigate, Enter confirms+moves down, Tab confirms+moves right, Escape cancels. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When typing to start editing, cursor is placed at end instead of selecting all, so the typed character is preserved. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Browser focus() auto-selects input text. Using requestAnimationFrame to set selection after focus settles, and a ref (not state) to track whether to select-all or cursor-at-end. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- TCPO search: each word matches descricao (accented) or search_term (stripped) - SINAPI search: per-word ilike on descricao, filter composicao only - SINAPI pricing: use custo_sem_desoneracao with fallback to custo_com_desoneracao - Remove codigo from search fields (description-only search) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- TCPO demolições: 57 compositions (67 search terms) - TCPO complementar: 552 compositions (142 new terms, all categories) - SINAPI sem desoneração: 7712 prices updated from Caixa Excel CSD sheet - SINAPI analítico: 38680 composition-insumo records imported - Fix search_term: strip accents, use codigo+descricao for all 1850 TCPO rows - Total TCPO: 1850 compositions, SINAPI analytical: 38680 records Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- New items inherit adm_percentual from project premissas (no hardcoded default) - New projects require adm% to be set explicitly - Excel export generates live formulas: SUM for parents, E+F for custo_total, SUMPRODUCT for administração Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds expandable chevron on each SINAPI/TCPO result row showing: - Material vs Mão de Obra cost breakdown summary - Full insumo table with código, descrição, classe, coeficiente, preço unitário, total - TCPO: colored badges (MOD/MAT/EQH) + LS+BDI split - SINAPI: fetches ob_sinapi_composicao_insumos with joined custo_sem_desoneracao Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Início do pivot do orcabot pra Orcamentista IA (SaaS de orçamento). Mantém: - frontend/ (Vite + React 18, páginas SINAPI/TCPO/Project/Dashboard) - supabase/migrations/ (28 migrations, schema ob_* multi-tenant) - scripts/ (data ingest SINAPI/TCPO — Python + TS scrapers) - .vercel/project.json (prj_GWZ4A9...) - .env.example, .gitignore, tsconfig, etc. Apaga: - src/ (NanoClaw Node runtime + LLM proxy + container runner) - agents/ (9 agents NanoClaw) - container/, groups/, launchd/, setup/, repo-tokens/, config-examples/ (infra NanoClaw) - tests/ (testes do runtime NanoClaw) - assets/ (branding NanoClaw) - .github/ (workflows NanoClaw: bump-version, ci, label-pr, update-tokens) - Dockerfiles, docker-compose, setup.sh - READMEs ja/zh, CHANGELOG, CONTRIBUTING, CONTRIBUTORS, skills-lock.json - vitest configs, .dockerignore, bun.lock (raiz) Próximo: criar api/index.py (FastAPI ASGI Vercel Python).
Sprint 0 — esqueleto da Engine Python para Orcamentista IA.
Adiciona:
- api/index.py — FastAPI app entry (ASGI, detectado pelo @vercel/python@5.x)
- api/routers/{ask,extract,dxf,sinapi}.py — stubs 501 (Sprints 1-3)
- api/lib/supabase.py — service-role client com lru_cache
- api/lib/auth.py — dual auth: Supabase JWT OU ORCABOT_API_SECRET (compat com frontend)
- api/lib/cache.py — helpers de hash p/ ob_vision_cache (Sprint 1)
- api/schemas/{ask,extract,dxf}.py — Pydantic placeholders (ask.RespostaOutput pronto)
- api/engines/{vision,dxf,sinapi}.py — placeholders Sprints 1-3
- vercel.json (raiz) — buildCommand frontend Vite + função Python (60s, 1024MB)
- requirements.txt — fastapi, uvicorn, gaik[vision-extract], ezdxf, openpyxl,
supabase-py, python-jose, sentence-transformers
- package.json (raiz) — minimal: dev:web, dev:api, build, typecheck
Atualiza:
- .env.example — vars Supabase/LLM/CORS/VITE_* documentadas
- .gitignore — adiciona Python (__pycache__, .venv) e remove regras NanoClaw
Apaga:
- tsconfig.json, eslint.config.js, .prettierrc (raiz) — apontavam pra src/ deletado;
frontend/ tem suas próprias configs
- .husky/pre-commit — husky desinstalado junto com runtime NanoClaw
Endpoints expostos:
- GET /api/healthz → 200 (smoke)
- POST /api/ask → 501 (Sprint 1)
- POST /api/extract → 501 (Sprint 2)
- POST /api/dxf/{...} → 501 (Sprint 2)
- POST /api/sinapi/match → 501 (Sprint 3)
Vercel preview vai validar:
- Frontend builda e sobe (Login page renderiza)
- /api/healthz retorna 200 com payload JSON
- Páginas que chamavam /api/process, /api/agent-chat, /api/caderno-query
vão dar 404 — esperado, migração no Sprint 1+.
Próximo: 0.9 inventário endpoints + 0.10 migration ob_vision_cache + 0.11 docs.
…e CLAUDE.md - docs/ARCHITECTURE.md — diagrama Vercel + Supabase, decisões, restrições, roadmap 5 sprints - docs/DEPLOY.md — projeto Vercel (prj_GWZ4A9...), env vars, smoke check, limites Hobby - docs/DEV.md — setup local Vite + uvicorn, dev flow, convenções de commit - docs/MIGRATION.md — mapeamento endpoints legados (/api/process, /api/agent-chat, /api/caderno-query) → novos (/api/ask, /api/extract, /api/sinapi/match) por página - supabase/migrations/20260410000002_vision_cache_and_queries.sql — arquivo da migration que foi aplicada via MCP apply_migration no remoto. Tabelas ob_vision_cache (chave: pdf_sha256+question_hash) e ob_vision_queries (audit/memorial). RLS via padrão ob_org_members.user_id = auth.uid(). - README.md raiz reescrito — Orcamentista IA (sem NanoClaw) - CLAUDE.md raiz reescrito — instruções pra futuras sessões Claude Code Sprint 0 completo. Próximo: abrir PR draft pra main, validar preview Vercel verde, parar para revisão antes de iniciar Sprint 1 (/api/ask).
Vercel project settings atualizados via API: - rootDirectory: 'frontend' → null (raiz do repo) - framework: 'vite' → null (vercel.json controla) - buildCommand/devCommand/installCommand/outputDirectory → null (vercel.json controla) Vercel agora consegue ler vercel.json raiz + ver pasta api/ pra publicar a Python function. Build command vem do vercel.json: 'cd frontend && bun install && bun run build'.
…t from .py + requirements.txt Vercel rejeitava 'Function Runtimes must have a valid version'. A sintaxe correta seria pinning exato (ex: '@vercel/python@5.0.0'). Mais simples: remover o campo runtime e deixar Vercel auto-detectar via extensao .py + requirements.txt na raiz.
Vercel deploy falhou: 'Total bundle size (5003.01 MB) exceeds Lambda ephemeral storage limit (500 MB)'. Causa: sentence-transformers puxa torch que puxa nvidia-cublas-cu12, nvidia-cudnn-cu12, etc — explode pra 5GB. Sprint 0 nao usa embedding mesmo (Sprint 3 task), entao removido por enquanto. Sprint 3 vai usar OpenAI text-embedding-3-small com dimensions=384, mantendo schema vector(384) atual. Re-embedar 13597 chunks single-shot ~$1.36, ~10min de runtime. Tambem pinando Python 3.13 via .python-version (era 3.12 default, 3.13 recomendado nas Vercel knowledge updates 2026-02-27).
github-actions
Bot
added
the
follows-guidelines
This was referenced May 15, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Type of Change
.claude/skills/<name>/, no source changes)Description
For Skills