Dev#370
Open
Chenglong-MS wants to merge 30 commits into
Open
Conversation
1. Add custom property support for height , label , and sx to AnvilLoader 2. Replace globally hardcoded loading text with customizable label parameter 3. Optimize loading overlay styles with new frosted glass background effect 4. Unify loading state display in App.tsx and VisualizationView
test: keep zh locale keys aligned
…support Introduce MiniAnalystAgent, a single-decision, low-cost variant of the analyst agent that emits the same frontend streaming event contract (visualize / explain) as the standard agent, so it is a drop-in replacement selectable from the UI. - analyst/mini_agent.py: MiniAnalystAgent (subclass of AnalystAgent). One turn: decide -> either "visualize" (data-transform code + chart spec, committed via the shared core-skill dispatch so action/result events are identical to the standard agent) or "explain" (free-form text surfaced via the frontend Pure-Q&A path). Optional one-shot data inspection; JSON action contract for small/local models. - routes/agents.py: agent_mode switch to swap AnalystAgent <-> MiniAnalystAgent. - src/app/App.tsx, dfSlice.tsx, SimpleChartRecBox.tsx: frontend mini-mode toggle, miniMode redux state, and agent_mode in the analyst streaming request body. - agents/client_utils.py: open-model (Ollama) enablement -- replay buffered responses as streaming chunks, salvage JSON tool calls emitted as plain content, and retry without reasoning_effort for models that lack "think" support. - analyst/agent.py: tolerate non-string chart encodings from weak models (repairable "field not found" instead of an unhashable-type crash). - analyst/__init__.py: export MiniAnalystAgent. - tests: test_mini_agent.py (23) and test_client_utils.py additions; all green. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The mini_notool variant (MiniAnalystAgent with allow_inspection=False: a no-tools, one-shot ablation used only in evaluation) is removed. MiniAnalystAgent now always offers the single optional inspection before committing its answer; the per-turn inspection-budget logic in _decide/_mini_tools is unchanged. Tests updated to drop the no-tool-variant cases.
Replace the long-form README with a condensed plan.md (high-level goal and steps). The detailed report (report.md) and working artifacts (work/) remain local-only and are intentionally untracked.
Mini analyst agent
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.
Made some good updates, but a little bit busy to write about it... so let's make the pr simple
This pull request introduces several significant improvements and cleanups to the Data Formulator agent codebase, focusing on agent configuration, chart restyling, and code explanation. It removes deprecated agents, adds robust validation for chart UI controls, and clarifies formula formatting in code explanations. The changes aim to improve reliability, maintainability, and security, especially around LLM-generated UI controls.
Agent configuration and cleanup:
data_agent,report_gen,experience_distill) with updated or renamed agents (analyst,workflow_distill) inagent_config.py, reflecting the current production agent roster.ChartInsightAgentfrom the codebase, further cleaning up the agent registry and code. [1] [2]Chart restyling and LLM UI controls:
agent_chart_restyle.pysystem prompt to support a newconfigUIfield, allowing LLMs to propose safe, declarative UI controls for chart variants. The prompt now includes detailed rules and examples for these controls._sanitize_config_ui, a robust validator that ensures LLM-authoredconfigUIcontrols are well-formed, have safe paths (no prototype pollution), and meet type requirements before returning them to the frontend. [1] [2]Formula formatting and code explanation:
Documentation and evaluation: