The transcripts are saved as is and viewed as is. Not parsed into prompt, tool calls, responses etc.
The schema for each agent transcript also changes and has changed in the past.
How to reliable parse?
One method is have a bank of schemas (to periodically added when necessary) for a coding agent and versioned with a manually mantained current version. A transcript tries to parse from current verion to all untill it finds one that parses correctly and then that transcript schema get locked and hence forth will always use that version of schema to be used.
The transcripts are saved as is and viewed as is. Not parsed into prompt, tool calls, responses etc.
The schema for each agent transcript also changes and has changed in the past.
How to reliable parse?
One method is have a bank of schemas (to periodically added when necessary) for a coding agent and versioned with a manually mantained current version. A transcript tries to parse from current verion to all untill it finds one that parses correctly and then that transcript schema get locked and hence forth will always use that version of schema to be used.