IssueFlow
IssueFlow is a learning-proposal management tool for agentic programming workflows, supporting agents in submitting learning proposals, a librarian agent in organizing and generalizing them, human approval, and management of rule retirement. Its core goal is to let agents accumulate reusable rules from failures, avoiding instruction pile-up and token waste.
Timeline (last 90 days)
IssueFlow is a workflow that lets AI coding agents propose learning proposals themselves, has a librarian agent organize and merge them, and then has humans approve and periodically clean up the rules.
In IssueFlow, agents can submit learning proposals, and the librarian agent is responsible for collecting, merging, and generalizing these proposals, organizing them into candidate rules with the actual instructions to be sent to the agent.
The librarian agent itself does not produce learning proposals and does not make final approval decisions, but rather performs the work of an intermediate processing layer.
When generalizing and merging proposals, the librarian agent processes them separately by agent, because the proposals come from a specific agent at a specific stage.
IssueFlow's review process allows humans to choose to apply a rule to a specific stage, or to make the rule a general rule.
Rule retirement is a decision process based on usage data: the librarian proactively proposes retirement based on usage history, and human operators can also manually retire rules.
When the author rejects a proposal with a reason, the reason is recorded, and the librarian will use it later when handling similar proposals.
The author provides an IssueFlow workflow demo showing a design approval and an unexpected decision from a real run.
All Facts (8)
IssueFlow is a workflow that lets AI coding agents propose learning proposals themselves, has a librarian agent organize and merge them, and then has humans approve and periodically clean up the rules.
50%UnverifiedIn IssueFlow, agents can submit learning proposals, and the librarian agent is responsible for collecting, merging, and generalizing these proposals, organizing them into candidate rules with the actual instructions to be sent to the agent.
50%UnverifiedThe librarian agent itself does not produce learning proposals and does not make final approval decisions, but rather performs the work of an intermediate processing layer.
50%UnverifiedWhen generalizing and merging proposals, the librarian agent processes them separately by agent, because the proposals come from a specific agent at a specific stage.
50%UnverifiedIssueFlow's review process allows humans to choose to apply a rule to a specific stage, or to make the rule a general rule.
50%UnverifiedRule retirement is a decision process based on usage data: the librarian proactively proposes retirement based on usage history, and human operators can also manually retire rules.
50%UnverifiedWhen the author rejects a proposal with a reason, the reason is recorded, and the librarian will use it later when handling similar proposals.
50%UnverifiedThe author provides an IssueFlow workflow demo showing a design approval and an unexpected decision from a real run.
50%