Spaces:
Sleeping
Sleeping
Commit History
-
fix: Cache-hit whitelist and LangGraph-dropped gate state
- Cache-hit path rebuilt the result from a hardcoded field list that predated metric_reference/recommendations/quality_notes; cache-served views therefore fell back to positional table refs misaligned with the SWOT text citations (the reported UI wonkiness). Full cached dict now passes through with only data_source/provider overridden. - data_gaps/suspect_metrics were not declared in AgentState, so LangGraph dropped them between researcher and analyzer - the gate's DATA NOT PROVIDED table section, scrubbing, and quality-notes surfacing were silently inert. Declared in the schema.096df90
-
Fix revision loop: add critique_details to AgentState
Bug: critique_details was not defined in AgentState TypedDict, causing LangGraph to drop it between nodes. Revisions never triggered because analyzer couldn't see critic's REJECTED status. - Add critique_details: Optional[dict] to AgentState - Add analyzer_revision_skipped: Optional[bool] to AgentState - Add debug logging to trace state flow Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>95ae0b4
-
Layer 1: Add metric reference table to prevent LLM hallucination
- Add immutable metric reference table at top of analyzer prompts - Use exact 'as of' dates (e.g., 2024-09-28) instead of ambiguous FY labels - Store metric_reference dict and SHA256 hash in state for integrity verification - Include reference table in both initial and revision prompts - Instruct LLM to copy values verbatim, not round or estimate New functions: - _format_metric_for_reference(): Format single metric with as-of date - _generate_metric_reference_table(): Build numbered reference table - _compute_reference_hash(): SHA256 hash for integrity check - _verify_reference_integrity(): Verify dict hasn't been corrupted Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>d2df6b8
-
Initial commit: Instant SWOT Agent
Multi-agent strategic analysis system with: - Research Gateway (A2A client to Researcher-Agent) - Analyzer, Critic, Editor workflow nodes - Data grounding enforcement in all LLM calls - React frontend with real-time progress 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>0c591a7