What happened?
After deploying Resonant Mind v3.2.0 cleanly to a Neon Postgres backend, standalone mind_write calls with type=observation against an existing entity return a generic "Internal tool error". The call does not write the observation. Same code path that was supposed to be fixed by v3.2.0 (silent observation-write failures) — now returns an error string instead of failing silently, but the write still does not land.
Adjacent mind_write paths work fine: entity creation with embedded observations, journal entries, all read tools. Only standalone observation writes against existing entities break.
What did you expect?
The observation should be written to the entity, vectorized, and made retrievable via mind_search. A success message should be returned, consistent with how observations land when embedded in entity-creation calls.
Steps to reproduce
Create an entity successfully:
mind_write({
type: "entity",
name: "test-entity",
entity_type: "threat-profile",
observations: ["Embedded observation in entity creation."],
source: "external",
certainty: "known",
tags: ["test"]
})
Result: Entity 'test-entity' created/updated with 1 observations (vectorized) ✓
Try to write a standalone observation against the now-existing entity:
mind_write({
type: "observation",
entity_name: "test-entity",
observations: ["Standalone observation test."],
source: "realization",
weight: "light",
certainty: "known",
tags: ["test"]
})
Result: Internal tool error
Confirm mind_search finds the embedded observation from step 1 but NOT the standalone observation attempted in step 2.
Environment
Resonant Mind version: v3.2.0
Storage backend: Neon Postgres (hand-migrated from the SQLite/D1 patch via 22 Postgres statements run in Neon SQL editor, all committed clean)
MCP client: Claude.ai (Cowork via Claude Agent SDK)
Wrangler version: Amanda — run npx wrangler --version and paste here
Error output
Internal tool error
No further detail in the MCP response. No visible stack trace, no SQL error surfaced. Worker logs may have more — check tailing npx wrangler tail against resonant-mind-cub while reproducing if you want richer context.
What happened?
After deploying Resonant Mind v3.2.0 cleanly to a Neon Postgres backend, standalone mind_write calls with type=observation against an existing entity return a generic "Internal tool error". The call does not write the observation. Same code path that was supposed to be fixed by v3.2.0 (silent observation-write failures) — now returns an error string instead of failing silently, but the write still does not land.
Adjacent mind_write paths work fine: entity creation with embedded observations, journal entries, all read tools. Only standalone observation writes against existing entities break.
What did you expect?
The observation should be written to the entity, vectorized, and made retrievable via mind_search. A success message should be returned, consistent with how observations land when embedded in entity-creation calls.
Steps to reproduce
Create an entity successfully:
mind_write({
type: "entity",
name: "test-entity",
entity_type: "threat-profile",
observations: ["Embedded observation in entity creation."],
source: "external",
certainty: "known",
tags: ["test"]
})
Result: Entity 'test-entity' created/updated with 1 observations (vectorized) ✓
Try to write a standalone observation against the now-existing entity:
mind_write({
type: "observation",
entity_name: "test-entity",
observations: ["Standalone observation test."],
source: "realization",
weight: "light",
certainty: "known",
tags: ["test"]
})
Result: Internal tool error
Confirm mind_search finds the embedded observation from step 1 but NOT the standalone observation attempted in step 2.
Environment
Resonant Mind version: v3.2.0
Storage backend: Neon Postgres (hand-migrated from the SQLite/D1 patch via 22 Postgres statements run in Neon SQL editor, all committed clean)
MCP client: Claude.ai (Cowork via Claude Agent SDK)
Wrangler version: Amanda — run npx wrangler --version and paste here
Error output
Internal tool error
No further detail in the MCP response. No visible stack trace, no SQL error surfaced. Worker logs may have more — check tailing npx wrangler tail against resonant-mind-cub while reproducing if you want richer context.