Repository navigation
Feature/experimental - #593
Open
leopiccionia wants to merge 419 commits into
Open
leopiccionia wants to merge 419 commits into
leopiccionia wants to merge 419 commits into
Conversation
…dary point limits
- Replace static default system prompt with dynamic calibration-aware builder
(AI_Adapter::get_calibration_aware_prompt) that reads active calibration
settings (granularity, confidence, title weight, thresholds, max points)
and injects them into the prompt when no custom prompt is used.
- Add adaptive precision instruction in build_enforced_schema() so the
confidence calibration rule does not conflict with the JSON schema block.
- Add new calibration controls: Primary Points and Secondary Points
(ai_cal_primary_max / ai_cal_secondary_max, 1-100, toggle-enabled).
UI uses single range sliders positioned below their respective thresholds.
- Propagate max-point rules through the AI Prompt Assistant meta-prompt
so generated custom prompts also respect the configured limits.
- Enforce max limits in post-processing:
- Frontend sidebar: applyPointLimits() truncates after threshold scoring
- Bulk processor: array_slice() after relevance classification
- Ensure full compatibility with NeuronAI Structured Output:
- Update Georeference_Result schema description to be neutral and not
contradict calibration filtering rules.
- Fix missing max-point rules in prompt generator calibration block.
- Remove duplicate boolean sanitization entries for new toggles.
- Update .architecture/ai/README.md to document calibration-aware prompt
and max-point enforcement.
Files changed:
- class-ai-adapter.php
- class-ai-handler.php
- class-jeo.php
- class-settings.php
- class-bulk-processor.php
- class-georeference-result.php
- tab-provider.php
- settings-page.js
- posts-sidebar/index.js
- .architecture/ai/README.md
Add convention entry explaining that the default AI system prompt is calibration-aware and listing the 5 locations that must be updated when adding new calibration controls (adapter, handler, frontend, bulk, schema).
Resolved conflict in class-georeference-result.php: - Kept remote's array() syntax for consistency - Kept HEAD's neutral schema description to avoid contradicting calibration filtering rules
…d robustness
- Replace hardcoded FileVectorStore topK=3 with configurable ai_rag_topk
setting (default 10, range 1-50) in JEO AI Settings > Knowledge Base.
RAG_Agent::vectorStore() reads the setting and falls back to 10.
- Add layer deduplication in RAG_Worker::on_layer_save():
Layer_Data_Loader now sets Document sourceType='layer' and
sourceName={post_ID}. Before adding on save, deleteBy() removes
previous embeddings for the same layer_id.
- Fix minimap-editor.js frontend issues:
- loadedLayers.length > 0 guard prevents empty-map flash during load
- Debounce onMove/onZoom (300ms) with lodash to reduce re-renders
- Simplify layerIds useMemo dependency from JSON.stringify to allLayers
All changes pass PHPCS and PHP syntax checks.
…nd robustness - Add RAG Vector Store Configuration section to .architecture/ai/README.md documenting ai_rag_topk setting and layer deduplication strategy. - Update .architecture/minimap/README.md conventions with debounced map interaction and layer render guard notes. - Add ai_rag_topk validation note to AGENTS.md Settings Validation.
… recreation
- Add validate_layers() and is_valid_layer() to Minimap class to filter
out layer IDs returned by the agent that no longer exist or are not
published map-layer CPTs. Prevents 'ghost' layers in the block.
- Also validate base_layer->id in run_agent() and null it if invalid.
- Remove Map key={key} and the useEffect that recreated the Map on
align/screen-width changes. The ResizeObserver in maplibregl-react
already handles container resize, so forced recreation was causing
unnecessary state loss (pan/zoom reset).
PHPCS clean, PHP syntax verified.
…d.sh Adds ensure_node_24() that: - Checks current node -v for v24 - Falls back to NVM (/home/diegosanches/.nvm, ~/.nvm, or shell function) - Auto-installs Node 24 via nvm if not present - Fails gracefully with clear message if neither works Prevents build failures when the shell default Node is not 24.
Introduces foundational infrastructure for converging all AI features onto the hacklabr/ai-assistant library: - JEO_AI_Factory: central Assistant builder with shared config, provider resolution, storage adapters, and tool wiring. - Tool_Registry: singleton registering all NeuronAI tools (search_layers, geocode, generate_layer, get_post_content). - AI_REST_Permissions: reusable permission callbacks replacing 12 inline closures in AI_Handler. - REST args schemas: added validation to ai-georeference, ai-test-key, ai-get-models, ai-test-retrieval, ai-clear-store, ai-delete-backup endpoints. - Minimap_Agent: now consumes tools via Tool_Registry instead of hardcoding class names. All new files pass PHPCS and php -l.
…rsations Phase 2 of the AI architecture convergence (minilayer excluded per request): - System_Prompt_Builder: extracts calibration-aware prompt construction, schema blocks, and prompt sanitization from AI_Adapter into a reusable, testable static class. AI_Adapter methods now delegate to the builder and are marked @deprecated. - AI_Handler::api_test_key: now uses System_Prompt_Builder::for_test_connection() instead of a hardcoded prompt string. - Georeferencing_Conversation: new class managing multi-turn georeferencing sessions via ConversationStore + WP_Storage. Supports start, user messages, assistant responses, thread trimming (max 20 messages), and session listing. - REST endpoint: POST /jeo/v1/ai-georeference-chat with full args schema (post_id, conversation_id, message, title, content). Enables future frontend chat refinement of AI-suggested locations. - Neuron_Adapter migration to JEO_AI_Factory: deferred to Phase 3 due to token-tracking compatibility risks. The infrastructure (factory, registry, builder) is now in place to support it safely. All files pass PHPCS and php -l. Production build verified.
Replaces the direct NeuronAI Agent extension with the unified hacklabr/ai-assistant Assistant model using declarative MCP config: - JEO_AI_Factory: added param to create_assistant() and new create_minilayer_assistant() method that passes Mapbox DevKit MCP config declaratively to AssistantConfig. - Minilayer_Agent: refactored from to a static factory that creates Assistant via JEO_AI_Factory. generate() now calls Assistant::chat() instead of Agent::chat(). MCP tools are discovered automatically by the library via McpConfigBridge. - Minilayer_Handler: updated to use AI_REST_Permissions::edit_posts() and added REST args schema for prompt + layer_name. - Minilayer_Service: updated to call Minilayer_Agent::generate() static method. No behavioral changes. - Generate_Layer_Tool: unchanged — still delegates to Minilayer_Service. - AGENTS.md + .architecture/ai/README.md + .architecture/minilayer/README.md: updated to document the unified architecture, new files, and MCP integration pattern. All files pass PHPCS and php -l. Production build verified.
PHPCBF auto-fix for spacing and alignment issues.
- Change ai_use_structured_output default from false to true - Add warning notice listing limitations (no token tracking, auto-fallback, internal tool bypass) - Add info notice with per-provider recommendations (OpenAI/Gemini/Anthropic = full support; Ollama/HuggingFace = variable) - Update AGENTS.md with structured output conventions, fallback behavior, Prompt Assistant interaction, and token tracking caveat
…ayer_Handler The file is in namespace Jeo\AI, so AI\AI_REST_Permissions resolved to Jeo\AI\AI\AI_REST_Permissions (non-existent). Use unqualified AI_REST_Permissions instead, which correctly resolves to Jeo\AI\AI_REST_Permissions. This was causing a fatal error on rest_api_init, breaking all REST endpoints.
Adds a Gutenberg sidebar plugin that suggests new paragraphs for posts using the site's RAG vector store (jeo_knowledge) and multi-turn AI chat. Backend: - Retrieve_Knowledge_Tool: semantic search against jeo_knowledge - Context_Generation_Output: structured output DTO (paragraphs, references) - Context_Agent: Assistant factory with post_analyzer sub-agent - Context_Handler: REST endpoints /context/setup, /chat, /state - Persist conversation state in post meta for session restoration - Customizable system prompt via ai_context_prompt setting Frontend: - contextSidebar webpack entry with PluginDocumentSettingPanel - Chat UI with textarea, expand modal, and suggestion cards - Inline linkify for referenced articles within paragraph text - Insert (core/paragraph block) and Copy (rich text) actions Docs: - Update .architecture/ai/README.md and AGENTS.md
…ettings - Add ai_use_context_custom_prompt checkbox to enable/disable custom prompt - Show built-in default prompt as read-only reference when custom is disabled - Toggle visibility between custom textarea and default reference via JS - Respect ai_use_context_custom_prompt flag in Context_Agent::system_prompt() - Update tab-context.php, class-settings.php, and class-context-agent.php
Enter submit the chat form. Shift+Enter add a line break
- New sections: navbar, hero, resource slider, features showcase, how it works, live demo, workshops, footer - PT-BR/EN i18n - Official Lottie animations + code-rebuilt ones - Hero background video - Performance optimizations - Mobile tab UX - Docs updated
…r fixes — issue #669 - Official photo masters for Mapas and layers: permuted, full-PNG render without mattes, 75% glass opacity on upper layers, shadow removal - Hero mobile order + 32px gap - Footer responsive below 1400px - Tab centering on auto-advance - Disabled workshop form button - i18n and docs updates
- VitePress 1.6 with brand theme (JEO Maps palette, Open Sans, light/dark logos) - Directory URL parity with MkDocs via rewrites + cleanUrls - Internal links migrated to absolute routes - docs/index.md redirects to the landing (React, outside VitePress) - docs-site.yml builds with Node and keeps the site/ bot commit - Remove mkdocs.yml and requirements-docs.txt
- Merge feat/landing-redesign (site-novo/www React landing) - build:site builds the landing into site/ (base /) and the docs into site/docs/; order matters because the landing build empties site/ - Landing docs links point to /docs/ - Remove docs pages that were not in the MkDocs nav (usability-issues) - docs-site.yml runs the single command and commits site/
Design references (site-novo/figma-ref, CLAUDE.md) stay in site-novo/. The landing build now outputs to ../site from the new location.
Works in every environment: local preview stays on localhost, production lands on jeowp.org/. Nav and sidebar Home links point at /docs/ (base applied) and follow the same redirect.
- Site name is JEO Maps; navbar shows the wordmark logo alone with alt text, 40px on mobile and 50px on desktop - Home links point at the landing page (same tab) - New Download item in the navbar - /docs/ redirects to /docs/index-plugin/ - Drop the wordpress.org compliance/audit dev pages
Also point the documentation CTAs at /docs/index-plugin/.
The landing dev server proxies /docs to the VitePress dev server so both halves of the site share one origin (http://localhost:5173) with live reload on each side.
This branch has not been deployed
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
3 participants
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.
No description provided.