After LLM returns scene IR, validate that critical fields (contentType,
responsePath, workflowArchetype) are present. If missing, send one
follow-up prompt to fill gaps. Merges repaired fields without overwriting
valid data from the first extraction.
🤖 Generated with [Qoder][https://qoder.com]
Identify and push js/ directory business logic files (mca.js, sgApi.js,
etc.) to the LLM prompt. Exclude third-party libraries. Capped at 4
fragments to stay within MAX_DEEP_PROMPT_CHARS budget.
🤖 Generated with [Qoder][https://qoder.com]
Explicitly require LLM to fill contentType, responsePath, and
requestTemplate with detected values or defaults. Reduces empty-field
rate from ~60% to target ~10%.
🤖 Generated with [Qoder][https://qoder.com]
When the LLM returns an empty modes array but has apiEndpoints,
automatically create a default mode with the first endpoint,
requestTemplate, responsePath, and normalizeRules. This ensures all
scenes compile through the multi-mode path.
🤖 Generated with [Qoder][https://qoder.com]
Add analyzeSceneDeep() function that extracts complete SceneInfo from
index.html content including API endpoints, static parameters, column
definitions, and business logic. Also adds http module import to support
non-HTTPS LLM endpoints.
🤖 Generated with [Qoder][https://qoder.com]
- Add /select-folder and /select-file APIs using PowerShell dialogs
- Add --target-url parameter to CLI for explicit target URL override
- Redesign Web UI with folder browse buttons for all path inputs
- Add target_url optional input field for specifying target page URL
- Auto-fill scene-id from selected folder name
🤖 Generated with [Qoder][https://qoder.com]
Add scene type selector to the generator form:
- CSS styles for select element matching existing input style
- Dropdown with "报表收集类" (report_collection) and "监测类" (monitoring) options
- Pass sceneKind to /generate API endpoint
🤖 Generated with [Qoder][https://qoder.com]
Pass sceneKind to sg_scene_generate CLI when specified,
allowing generation of different scene kinds (report_collection,
monitoring, etc.).
🤖 Generated with [Qoder][https://qoder.com]
Add parent sgClaw directory to config search candidates so Node.js
server automatically finds the correct config with DeepSeek API
endpoint instead of the test config in claw-new directory.
🤖 Generated with [Qoder][https://qoder.com]
- server.js: sanitize static file paths to prevent directory traversal
(GET /../../sgclaw_config.json would expose API key)
- config-loader.js: fix normalizeBaseUrl to strip /v1 before appending,
preventing double /v1 for non-standard base URLs
🤖 Generated with [Qoder][https://qoder.com]
Add self-contained HTML page for the Scene Skill Generator frontend:
- Dual-column glass-morphism layout matching service-console style
- Left sidebar: status card, sourceDir input with analyze button,
sceneId/sceneName inputs, settings button, generate button
- Right panel: streaming log display with SSE event rendering
- Settings modal: outputRoot, lessons, llmBaseUrl, llmModel fields
- JavaScript: connects to http://127.0.0.1:3210, implements analyze()
via fetch POST /analyze, generate() via fetch POST /generate with
SSE stream reading, settings modal open/close
- Rust test verifying HTML file exists and contains required elements
🤖 Generated with [Qoder][https://qoder.com]
Add config-loader.js module for loading sgclaw_config.json credentials
and resolving project root directory. Add initial Rust source-guard test
to verify server file paths exist.
🤖 Generated with [Qoder][https://qoder.com]
