agentdb-cypher

Cypher Query

Run arbitrary Cypher against the AgentDB graph backend. The native binding accepts standard openCypher; AgentDB also adds a few extensions (uplift, confidence properties on edges).

When to use

• Need a query the standard agentdb_causal_* tools don't cover
• Multi-hop traversal with custom filtering
• Aggregation queries — count nodes by type, find hub nodes, etc.
• Migration / inspection during debugging

API

agentdb_causal_query(
  cypher: <query string>,
  params?: { ... }
)

Returns: { rows, stats: { execMs, rowsMatched, rowsReturned } }

Example queries

-- Top 10 most-cited skills
MATCH (s:Skill)<-[r:references]-(:Episode)
RETURN s.name, count(r) AS cites
ORDER BY cites DESC LIMIT 10

-- Find hub nodes (high in-degree)
MATCH (n)<-[r]-()
RETURN n.id, n.type, count(r) AS inDegree
ORDER BY inDegree DESC LIMIT 20

-- Walk supersedes chain backwards from current ADR
MATCH p=(current:ADR {id: 'ADR-046'})-[:supersedes*]->(ancestor:ADR)
RETURN [n IN nodes(p) | n.title]

-- Cycle detection (worth a warning)
MATCH p=(n)-[:caused*1..5]->(n)
RETURN p LIMIT 5

Safety

The agentdb_causal_query tool accepts arbitrary Cypher, including writes (CREATE, DELETE, DETACH DELETE). For destructive operations, prefer the typed delete tools (agentdb_causal_node_delete, agentdb_causal_edge_delete) — they enforce cascade semantics and write to the attestation log.

Don't

• Don't use string interpolation for ids — use params to avoid Cypher injection.
• Don't run MATCH (n) DETACH DELETE n (deletes everything). Always scope the match.
• Don't loop a Cypher query in client code when one query with WHERE id IN [...] works — the binding's per-call overhead adds up.