knowledge-search

name	knowledge-search
description	Search the financial knowledge base for entities, events, relationships, timelines, and risk assessments. Use when the user asks about a company, person, or financial topic that requires structured knowledge retrieval.
allowed-tools	Bash
argument-hint	["entity names or query"]

Knowledge Search

Search the financial knowledge base via knowledge-cli search.

Parameters

Parameter	Type	Default	Description
--entities	list[str]	[]	Entity names, e.g. "小米集团" "苹果公司"
--intent	str	ENTITY_OVERVIEW	Query intent (see below)
--time-range	str	None	START:END in ISO dates, e.g. 2025-04-01:2026-04-13
--event-types	list[str]	None	Filter by event types, e.g. "债务违约" "股权质押"
--hops	int	1	Graph expansion hop count (1-5)
--target-entity	str	None	Second entity for A-B relationship queries
--top-k	int	20	Max results to return
--graph-enabled / --no-graph	flag	enabled	Toggle knowledge graph expansion

Intent Types

• ENTITY_OVERVIEW — General entity overview (default)
• ENTITY_TIMELINE — Entity timeline analysis
• RELATIONSHIP_QUERY — Relationship between two entities (requires --target-entity)
• COMPARATIVE_ANALYSIS — Compare multiple entities
• EVENT_ANALYSIS — Analyze specific events
• RISK_ASSESSMENT — Risk assessment for entities
• GUARANTEE_ANALYSIS — Guarantee relationship analysis
• TOPIC_RESEARCH — Research a topic across entities
• EVENT_IMPACT_ANALYSIS — Event impact analysis

Usage

Construct the command based on the user's request:

knowledge-cli search --entities "小米集团"

knowledge-cli search --entities "小米集团" --time-range 2025-04-01:2026-04-13 --intent ENTITY_TIMELINE

knowledge-cli search --entities "小米集团" --intent RELATIONSHIP_QUERY --target-entity "苹果公司" --hops 3

knowledge-cli search --entities "小米集团" "腾讯" --intent COMPARATIVE_ANALYSIS

Output

Returns JSON with:

• knowledge_units — Matched evidence units (summaries, entities, evidence text)
• entities — Matched entity records from text retrieval
• event_clusters — Aggregated event views from text retrieval
• graph_data — Graph metadata:
  • summary — Start entities, cluster count, entity count
  • clusters_overview — Lightweight cluster list (id, title, type, member_count, neighbor_entities, hit_reasons). Use this to decide which clusters to explore further.
  • nodes/edges/paths — Only present for relationship path queries (RELATIONSHIP_QUERY)
• retrieval — BM25 search metadata and scores
• errors — Any soft failures

Tier-2 Graph Expansion

When you need full graph detail for specific clusters, use:

knowledge-cli graph-expand --cluster-ids "clu_abc123" "clu_def456"

Returns JSON with nodes, edges, paths for the selected clusters.

Instructions

1. Parse the user's request to determine the appropriate --intent, --entities, and optional filters.
2. Run the command using PYTHONIOENCODING=utf-8 knowledge-cli search ... to ensure consistent encoding across systems.
3. The output is JSON printed to stdout with all non-ASCII characters escaped (\uXXXX). Parse it and present the key findings to the user.
4. If the user's query mentions two entities and their relationship, use --intent RELATIONSHIP_QUERY --target-entity "B" --hops N.
5. If the user asks about time progression, use --intent ENTITY_TIMELINE with a --time-range.
6. Entity names containing spaces do NOT need extra quoting beyond the surrounding quotes.
7. When the user asks about specific clusters or wants deeper analysis, note the cluster_id values from graph_data.clusters_overview and run knowledge-cli graph-expand --cluster-ids <ids> to get full nodes, edges, and paths.

Examples from user arguments

If the user invokes /knowledge-search 小米集团, run:

knowledge-cli search --entities "小米集团"

If the user invokes /knowledge-search 小米和腾讯的关系, determine intent and entities, then run:

knowledge-cli search --entities "小米集团" --intent RELATIONSHIP_QUERY --target-entity "腾讯" --hops 2