Agent Context / Fragmentation Map
Local evidence only
no provider claims
no provider claims
Where does the goal lose its memory?
The post describes separate tabs, workspaces, memories, and logins as the scattered mess an Agent OS should remove. This instrument measures that boundary in your own ordered workflow evidence.
Paste a workflow inventory. See the exact handoffs where context must be re-entered before you decide whether shared memory would help.
Evidence input
One row per workflow step
Loading local parser and map renderer
Use your own export or load the bundled sample.Saved locally
Measured result
Ready for evidence
No rows parsed
-rows
-handoffs
-memory re-entry
-changed dimensions
Run the analysis to draw each step and the context boundary between it and the next step.
The ratio is changed agent, workspace, and memory dimensions divided by three dimensions per adjacent handoff. It is a property of these supplied rows, not a model-quality score.
Read the boundary
Friction is a sequence, not a vibe.
Each card names the surfaces present in one workflow. A bright transition means the next step changed the context the agent must carry. The same-surface counterexample stays quiet by design.
HandoffAdjacent steps within one workflow.
Re-entryMemory surface changed or became none.
RatioChanged dimensions / possible dimensions.
Waiting for rows
Load the sample or paste a CSV to inspect the evidence.