diff --git a/CHANGELOG.md b/CHANGELOG.md index 2c8ed0ae..ffa0128c 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,6 +1,17 @@ # Changelog ## [Unreleased] +### Fixed + +- **LLM Schema Compliance**: Improved prompts to prevent LLMs from returning objects instead of plain strings for `list[str]` fields + - All graph prompts now explicitly state that list fields must contain plain strings only + - Added missing `query` and `confidence` fields to search agent output format documentation + - Fixes validation errors with less capable models that ignore JSON schema constraints +- **AG-UI Frontend Types**: Fixed TypeScript interfaces in ag-ui-research example to match backend Python models + - `EvaluationResult`: `confidence` → `confidence_score`, `should_continue` → `is_sufficient`, `gaps_identified` → `gaps`, `follow_up_questions` → `new_questions`, added `key_insights` + - `ResearchReport`: `question` → `title`, `summary` → `executive_summary`, `findings` → `main_findings`, removed `insights_used`/`methodology`, added `limitations`/`recommendations`/`sources_summary` + - Updated Final Report UI to display new fields (Recommendations, Limitations, Sources) + ## [0.20.1] - 2025-12-11 ### Added diff --git a/examples/ag-ui-research/frontend/components/Agent.tsx b/examples/ag-ui-research/frontend/components/Agent.tsx index a8dc5f59..8a98a3dd 100644 --- a/examples/ag-ui-research/frontend/components/Agent.tsx +++ b/examples/ag-ui-research/frontend/components/Agent.tsx @@ -53,20 +53,22 @@ interface ResearchContext { } interface EvaluationResult { - confidence: number; + key_insights: string[]; + new_questions: string[]; + gaps: string[]; + confidence_score: number; + is_sufficient: boolean; reasoning: string; - should_continue: boolean; - gaps_identified: string[]; - follow_up_questions: string[]; } interface ResearchReport { - question: string; - summary: string; - findings: string[]; + title: string; + executive_summary: string; + main_findings: string[]; conclusions: string[]; - insights_used: string[]; - methodology: string; + limitations: string[]; + recommendations: string[]; + sources_summary: string; } interface ResearchState { diff --git a/examples/ag-ui-research/frontend/components/StateDisplay.tsx b/examples/ag-ui-research/frontend/components/StateDisplay.tsx index b1bb2c39..c0ad3291 100644 --- a/examples/ag-ui-research/frontend/components/StateDisplay.tsx +++ b/examples/ag-ui-research/frontend/components/StateDisplay.tsx @@ -58,20 +58,22 @@ interface ResearchContext { } interface EvaluationResult { - confidence: number; + key_insights: string[]; + new_questions: string[]; + gaps: string[]; + confidence_score: number; + is_sufficient: boolean; reasoning: string; - should_continue: boolean; - gaps_identified: string[]; - follow_up_questions: string[]; } interface ResearchReport { - question: string; - summary: string; - findings: string[]; + title: string; + executive_summary: string; + main_findings: string[]; conclusions: string[]; - insights_used: string[]; - methodology: string; + limitations: string[]; + recommendations: string[]; + sources_summary: string; } interface ResearchState { @@ -179,7 +181,7 @@ export default function StateDisplay({ state }: StateDisplayProps) { state.max_iterations > 0 ? (state.iterations / state.max_iterations) * 100 : 0; - const confidence = state.last_eval?.confidence || 0; + const confidence = state.last_eval?.confidence_score || 0; return (
- {state.result.question} + {state.result.title}

- Summary + Executive Summary

- +
@@ -1072,7 +1074,7 @@ export default function StateDisplay({ state }: StateDisplayProps) { marginBottom: "0.5rem", }} > - Key Findings + Main Findings
-
-

- Methodology -

-
- + {state.result.recommendations.length > 0 && ( +
+

+ Recommendations +

+
    + {state.result.recommendations.map((rec, idx) => ( +
  • + +
    • + ))} +
-
+ )} + {state.result.limitations.length > 0 && ( +
+

+ Limitations +

+
    + {state.result.limitations.map((lim, idx) => ( +
  • + +
    • + ))} +
+
+ )}

- Insights Used ({state.result.insights_used.length}) + Sources

- {state.result.insights_used.map((insightId, idx) => { - const insight = state.context.insights.find( - (i) => i.id === insightId, - ); - return ( -
- {insight ? ( -
- -
- ) : ( -
- Insight ID: {insightId} -
- )} -
- ); - })} +
diff --git a/haiku_rag_slim/haiku/rag/graph/common/prompts.py b/haiku_rag_slim/haiku/rag/graph/common/prompts.py index 2f17ad7f..ed8d49c2 100644 --- a/haiku_rag_slim/haiku/rag/graph/common/prompts.py +++ b/haiku_rag_slim/haiku/rag/graph/common/prompts.py @@ -10,13 +10,17 @@ Responsibilities: Plan requirements: - Produce at most 3 sub_questions that together cover the main question. +- sub_questions must be a list of plain strings, where each string is a complete + question. Do NOT use objects with nested fields like {question, details}. - Each sub_question must be a standalone, self-contained query that can run without extra context. Include concrete entities, scope, timeframe, and any qualifiers. Avoid ambiguous pronouns (it/they/this/that). - Prioritize the highest-value aspects first; avoid redundancy and overlap. - Prefer questions that are likely answerable from the current knowledge base; if coverage is uncertain, make scopes narrower and specific. -- Order sub_questions by execution priority (most valuable first).""" +- Order sub_questions by execution priority (most valuable first). + +Use the gather_context tool once on the main question before planning.""" SEARCH_AGENT_PROMPT = """You are a search and question-answering specialist. @@ -46,8 +50,13 @@ Each result includes: - Type: content type like paragraph, table, code, list_item (when available) - Content: the actual text -IMPORTANT: In cited_chunks, use the EXACT, COMPLETE chunk ID (the full UUID). -Do NOT truncate or shorten chunk IDs. +Output format: +- query: Echo the question you are answering +- answer: Your concise answer based on the retrieved content +- cited_chunks: List of plain strings containing only the chunk UUIDs (not objects) +- confidence: A score from 0.0 to 1.0 indicating answer confidence + +IMPORTANT: Use the EXACT, COMPLETE chunk ID (full UUID). Do NOT truncate IDs. Guidelines: - Base answers strictly on retrieved content - do not use external knowledge. diff --git a/haiku_rag_slim/haiku/rag/graph/deep_qa/prompts.py b/haiku_rag_slim/haiku/rag/graph/deep_qa/prompts.py index 56d4d4d9..9dc46615 100644 --- a/haiku_rag_slim/haiku/rag/graph/deep_qa/prompts.py +++ b/haiku_rag_slim/haiku/rag/graph/deep_qa/prompts.py @@ -9,8 +9,10 @@ Task: - Be clear, accurate, and well-structured Output format: +- query: Echo the original question being answered - answer: The complete answer to the original question (2-4 paragraphs) -- cited_chunks: List of chunk IDs (from sub-answers) that directly support your answer +- cited_chunks: List of plain strings containing chunk IDs (UUIDs only, not objects) +- confidence: A score from 0.0 to 1.0 indicating answer confidence Guidelines: - Start directly with the answer - no preamble like "Based on the research..." @@ -30,7 +32,7 @@ Task: Output format: - is_sufficient: Boolean indicating if we can answer the question comprehensively - reasoning: Clear explanation of your assessment -- new_questions: List of specific follow-up questions needed (empty if sufficient) +- new_questions: List of plain strings, each a specific follow-up question (not objects) Guidelines: - Be strict but reasonable in your assessment diff --git a/haiku_rag_slim/haiku/rag/graph/research/prompts.py b/haiku_rag_slim/haiku/rag/graph/research/prompts.py index 540019ee..70dfbb22 100644 --- a/haiku_rag_slim/haiku/rag/graph/research/prompts.py +++ b/haiku_rag_slim/haiku/rag/graph/research/prompts.py @@ -16,19 +16,20 @@ Tasks: Output format (map directly to fields): - highlights: list of insights with fields {summary, status, supporting_sources, originating_questions, notes}. Use status one of {validated, open, tentative}. + supporting_sources and originating_questions must be lists of plain strings. - gap_assessments: list of gaps with fields {description, severity, blocking, resolved, resolved_by, supporting_sources, notes}. Severity must be one of - {low, medium, high}. resolved_by may reference related insight summaries if no - stable identifier yet. -- resolved_gaps: list of identifiers or descriptions for gaps now closed. -- new_questions: up to 3 standalone, specific sub-questions (no duplicates with - existing ones). + {low, medium, high}. resolved_by and supporting_sources must be lists of plain strings. +- resolved_gaps: list of plain strings (identifiers or descriptions for gaps now closed). +- new_questions: list of plain strings, up to 3 standalone questions (no duplicates). - commentary: 1–3 sentences summarizing what changed this round. +All list fields must contain plain strings only, not objects. + Guidance: - Be concise and avoid repeating previously recorded information unless it changed materially. -- Tie supporting_sources to the evidence used; omit if unavailable. +- For supporting_sources, use only the document_uri strings from the sources. - Only propose new sub_questions that directly address remaining gaps. - When marking a gap as resolved, ensure the rationale is clear via resolved_by or notes.""" @@ -58,15 +59,15 @@ Strictness: - Treat unresolved high-severity or blocking gaps as a hard stop. Output fields must line up with EvaluationResult: -- key_insights: concise bullet-ready statements of the most decision-relevant - insights (cite status if helpful). -- new_questions: follow-up sub-questions (max 3) meeting the specificity rules. -- gaps: list remaining blockers; reuse wording from the tracked gaps when - possible to aid downstream reconciliation. +- key_insights: list of plain strings, concise bullet-ready statements. +- new_questions: list of plain strings, follow-up sub-questions (max 3). +- gaps: list of plain strings, remaining blockers (reuse wording from tracked gaps). - confidence_score: numeric in [0,1]. - is_sufficient: true only when no blocking gaps remain. - reasoning: short narrative tying the decision to evidence coverage. +All list fields must contain plain strings only, not objects. + Remember: prefer maintaining continuity with the structured context over introducing new terminology.""" @@ -82,16 +83,13 @@ Goals: Report guidelines (map to output fields): - title: concise (5–12 words), informative. - executive_summary: 3–5 sentences summarizing the overall answer. -- main_findings: 4–8 one‑sentence bullets; each reflects evidence from the - research (do not include inline citations or snippet text). -- conclusions: 2–4 bullets that follow logically from findings. -- recommendations: 2–5 actionable bullets tied to findings. -- limitations: 1–3 bullets describing key constraints or uncertainties. -- sources_summary: List specific sources used with document paths, page numbers, - and section headings where available. Format each as: - "- /path/to/document.pdf (p. 5, Section: Introduction)" or - "- /path/to/file.md (Section: Getting Started)" - Include one bullet per distinct source document. +- main_findings: list of plain strings, 4–8 one‑sentence bullets reflecting evidence. +- conclusions: list of plain strings, 2–4 bullets following logically from findings. +- recommendations: list of plain strings, 2–5 actionable bullets tied to findings. +- limitations: list of plain strings, 1–3 bullets describing constraints or uncertainties. +- sources_summary: single string listing sources with document paths and page numbers. + +All list fields must contain plain strings only, not objects. Style: - Base all content solely on the collected evidence.