fix: scope summaries retrieval to actor for cross-session memory recall

[notgitika]

Summary

• Remove /{session_id} from the summaries namespace path in the generated memory/session.py template
• Fixes all three protocol variants (HTTP, A2A, AGUI) so agents with --memory longAndShortTerm can retrieve conversation summaries across sessions via hierarchical prefix matching
• Updates asset snapshots to reflect the template change

Problem

The generated retrieval config scoped summaries to /summaries/{actor_id}/{session_id}, meaning the agent could only recall summaries from the current session. This completely defeated the purpose of long-term memory for returning users.

Fix

Change the retrieval path from /summaries/{actor_id}/{session_id} to /summaries/{actor_id}. The SDK's retrieve_memories() uses hierarchical prefix matching, so this now finds summaries stored under any session for the actor.

Writes still store under /summaries/{actor_id}/{session_id} (SDK behavior), so session-level organization is preserved — only the read scope is widened.

Test plan

• npm run test:update-snapshots — snapshots regenerated
• Asset snapshot tests pass (109/109)
• Pre-commit hooks pass (typecheck, lint, secretlint)
• Manual verification: create agent with --memory longAndShortTerm, invoke across two sessions, confirm second session retrieves first session's summaries

Closes #1279

[github-actions]

Package Tarball

aws-agentcore-0.12.1.tgz

How to install

npm install https://github.com/aws/agentcore-cli/releases/download/pr-1299-tarball/aws-agentcore-0.12.1.tgz

[agentcore-cli-automation]

Fix looks correct and consistent with how retrieve_memories does hierarchical prefix matching. The write-side namespace (/summaries/{actorId}/{sessionId}) configured on the SUMMARIZATION strategy is unchanged, so existing data still organizes by session — only the read scope widens. Nice touch that this also brings create-time templates in line with the import translators (langgraph-translator.ts, strands-translator.ts), which already used /summaries/{user_id}/.

One thing worth addressing before merging:

docs/memory.md (line 95) still shows the buggy pattern this PR is fixing:

f"/summaries/{actor_id}/{session_id}": RetrievalConfig(top_k=3, relevance_score=0.5)

Since that section is the canonical "set up memory by hand" example, leaving it as-is will steer users toward the same bug the PR fixes in the templates. Suggest updating it to f"/summaries/{actor_id}" to match, with a short note explaining the asymmetry (writes go under /summaries/{actor_id}/{session_id}, reads use prefix matching for cross-session recall).