The Agent Lock-In Test Is A Full Session Export

Agent portability requires replayable messages, tools, evidence, compaction, subagent state, artifacts, versions, and a real delete path.

Retrieval answer

A transcript is not a portable agent session. The practical test is whether another runtime can import the full evidence and artifact bundle, explain previous decisions, continue the task, and delete the state without provider-only IDs.

New Runtime synthesiseditorial-diagram
A New Runtime whiteboard portability test moving a complete agent-session package between two runtimes and validating replay and deletion.
New Runtime synthesis from Earendil's session portability essay.New Runtime synthesisOriginal source ->

Field note

Conversation history is no longer the same thing as an agent session. Long-running work accumulates reasoning artifacts, hosted-search evidence, compaction, subagent messages, file and vector-store references, container state, and caches. If those remain behind provider IDs, an export moves only the outer transcript.

Earendil offers a useful boundary: a response ID is not a transcript, ciphertext without a user-controlled key is not user-controlled state, and a list of citations is not the retrieved passages and ranking the model actually saw. This state is better described as provider-sealed.

A portable export should contain messages, tool calls and complete results, search queries and retrieved passages with timestamps, readable compaction summaries, the subagent tree, an artifact manifest, prompt-tool-model versions, and file references that resolve outside the source platform.

The acceptance test is replay. Load the bundle into a second runtime, ask it to explain prior decisions, continue the task, and delete the imported state. Opaque blobs, dead file IDs, or a missing evidence chain are portability failures. Provider-native sealed state may improve continuation inside one ecosystem, but it should sit beside a readable handoff rather than replace it.

Recommendation

Agent portability requires replayable messages, tools, evidence, compaction, subagent state, artifacts, versions, and a real delete path.

Discovery graph / next reads

Continue through New Runtime

Open the graph
  1. 01topicAgent Memory - New RuntimeExplore the agent-memory topic hub.
  2. 02topicPortability - New RuntimeExplore the portability topic hub.
  3. 03topicAi Infrastructure - New RuntimeExplore the ai-infrastructure topic hub.
  4. 04archiveField NotesOpen the latest editorial analysis.
  5. 05source ledgerSource LedgerInspect the public source evidence graph.

These links are also published in this page's JSON twin and as typed edges in DiscoveryGraph v1.

Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract