personal memory agent
Fix entity extraction prompt to prevent literal 'type:' output
The LLM was interpreting "* type: name - description" literally,
outputting "* type: Person - Name - ..." which broke parsing.
Changed to explicit format with examples:
* Type: Entity Name - Description
Example:
* Person: Alice Smith - Mentioned in Slack...
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
+6
-2
+6
-2
think/insights/entities.md
+6
-2
think/insights/entities.md
···
27
27
Skip entities that don't fit one of these four types.
28
28
Skip any url, domain, filename, or path.
29
29
30
-
Output as a simple markdown list of detected entities with all three elements: the chosen type, the detected entity name, and the description of how the entity occurred in the document:
31
-
* type: name - description of how they appear across the segment (transcript or screen? which app? what do we learn about them?)
30
+
Output as a markdown list. Each line has three parts separated by colon and dash:
31
+
* Type: Entity Name - Description
32
+
33
+
Example:
34
+
* Person: Alice Smith - Mentioned in Slack discussing the project timeline
35
+
* Tool: Grafana - Visible on screen showing metrics dashboards