I’d like to propose an optional CogniCore memory integration for Knowledge Work Plugins.
Knowledge Work Plugins provide reusable workflows and instructions for common tasks. I think there is an interesting complementary layer missing:
What if the plugin could learn from previous completed tasks and reuse that experience later?
For example, a plugin handles a recurring workflow:
Task
↓
Approach A → failed
Approach B → failed
Approach C → successful
↓
Outcome verified
↓
Experience stored
On a future task, instead of starting completely from scratch:
New task
↓
Retrieve relevant experience
↓
Check whether the previous conditions still apply
↓
Use the previous experience as a starting point
↓
Verify the result
The important distinction is that this isn't intended to replace the plugin's instructions.
The plugin provides the workflow.
Memory provides experience accumulated from using the workflow.
What could be remembered?
For example:
Previous approaches that worked
Failed approaches and why they failed
User/project preferences
Successful outcomes
Verification evidence
Relevant environment/context
Provenance and timestamps
Experiences that have been superseded
I'd also like failed experiences to be first-class:
"This approach failed under these conditions."
That can be extremely valuable for preventing an agent from repeatedly making the same mistake.
Verification
I don't think an agent saying "this worked" should automatically become trusted memory.
The model I'm experimenting with is:
Observed
↓
Evidence
↓
Verified
↓
Promoted
↓
Reusable
↓
Revalidated in new context
This is particularly important for knowledge-work workflows because recommendations and processes can become outdated.
I've been developing CogniCore, an open-source cognitive infrastructure project for AI agents, around persistent memory, reflection, experience reuse, and transferable agent experience.
https://github.com/cognicore-dev/cognicore-my-openenv
I'd be interested in feedback from people working on Knowledge Work Plugins:
Would persistent, verified experience make these plugins more useful over repeated workflows, or is it better to keep the plugins deterministic and stateless?
I’d also be happy to prototype this against one existing Knowledge Work Plugin as a concrete demonstration.
I’d like to propose an optional CogniCore memory integration for Knowledge Work Plugins.
Knowledge Work Plugins provide reusable workflows and instructions for common tasks. I think there is an interesting complementary layer missing:
What if the plugin could learn from previous completed tasks and reuse that experience later?
For example, a plugin handles a recurring workflow:
Task
↓
Approach A → failed
Approach B → failed
Approach C → successful
↓
Outcome verified
↓
Experience stored
On a future task, instead of starting completely from scratch:
New task
↓
Retrieve relevant experience
↓
Check whether the previous conditions still apply
↓
Use the previous experience as a starting point
↓
Verify the result
The important distinction is that this isn't intended to replace the plugin's instructions.
The plugin provides the workflow.
Memory provides experience accumulated from using the workflow.
What could be remembered?
For example:
Previous approaches that worked
Failed approaches and why they failed
User/project preferences
Successful outcomes
Verification evidence
Relevant environment/context
Provenance and timestamps
Experiences that have been superseded
I'd also like failed experiences to be first-class:
"This approach failed under these conditions."
That can be extremely valuable for preventing an agent from repeatedly making the same mistake.
Verification
I don't think an agent saying "this worked" should automatically become trusted memory.
The model I'm experimenting with is:
Observed
↓
Evidence
↓
Verified
↓
Promoted
↓
Reusable
↓
Revalidated in new context
This is particularly important for knowledge-work workflows because recommendations and processes can become outdated.
I've been developing CogniCore, an open-source cognitive infrastructure project for AI agents, around persistent memory, reflection, experience reuse, and transferable agent experience.
https://github.com/cognicore-dev/cognicore-my-openenv
I'd be interested in feedback from people working on Knowledge Work Plugins:
Would persistent, verified experience make these plugins more useful over repeated workflows, or is it better to keep the plugins deterministic and stateless?
I’d also be happy to prototype this against one existing Knowledge Work Plugin as a concrete demonstration.