Skip to content

LEARNINGS.md Sync Gap

Problem

Learning pipeline writes to ge-ops/master/agent-configs/{name}/LEARNINGS.md. Executor reads from ge-ops/agents/{name}/LEARNINGS.md. Different paths.

Victoria has 340 lines of real security learnings that agents never see at runtime. The entire learning pipeline produces output that never enters the agent prompt.

The pitfalls doc (agent-system.md line 51-54) documents this mismatch but calls ge-ops/agents/ authoritative — that is where the empty stubs are.

Impact

All agent learnings since Jan 24 are invisible at runtime. Defeats the purpose of the knowledge layer.

Proposed Solution

Option A (preferred): Change identity loader to read from ge-ops/master/agent-configs/{name}/LEARNINGS.md. One-line change in ge_agent/identity/loader.py. Option B: Sync script copying between locations.

Acceptance Criteria

  • Executor loads real LEARNINGS.md content
  • Victoria gets her 340-line security learnings in prompt
  • Pitfalls doc updated