feat(memory-tools): resolve user from ADK tool_context

[nkanu17]

Closes #23

Memory tools now resolve the acting user in four steps: explicit user_id argument, then the invocation user from the ADK tool_context (its public user_id property), then config.default_owner_id, then config.default_user_id.

ADK already passes tool_context into run_async; the tools previously ignored it, which forced per-turn tool/agent/runner construction in multi-user workers just to scope memory via config.default_user_id. With this change a process-wide Agent and Runner can serve all users with correct owner scoping.

Implementation notes:

• The context read is defensive (getattr in try/except, empty and non-string values count as absent); a broken context never crashes a tool.
• Updated tools: create, search, update, delete, prompt. get.py fetches by memory id only and does not resolve a user, so it is unchanged.
• Backwards compatible: explicit arguments and config defaults behave exactly as before.
• Tests cover the full resolution order plus end-to-end scoping of search filters and create ownerId to the context user.

make check passes fully (format, lint, mypy, 119 passed / 12 skipped).

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 29e5741342

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

This PR updates the adk-redis memory tool suite to resolve the acting user from the ADK tool_context when no explicit user_id is provided, enabling safe reuse of a process-wide Agent and Runner in multi-user workers while preserving existing config-default behavior.

Changes:

• Added defensive tool_context.user_id handling to the shared _get_user_id resolution order.
• Plumbed tool_context through memory tool run_async implementations (create, search, update, delete, prompt).
• Added tests for user resolution and end-to-end scoping behavior, and documented the new resolution order.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/tools/test_memory_tools.py	Adds coverage for user resolution order and tool_context-based scoping in tool execution.
src/adk_redis/tools/memory/_base.py	Implements the new _get_user_id(..., tool_context=...) resolution order with defensive context reads.
src/adk_redis/tools/memory/create.py	Uses resolved user (including tool_context) when creating memories and stamping ownerId.
src/adk_redis/tools/memory/search.py	Uses resolved user (including tool_context) to scope search filters by ownerId.
src/adk_redis/tools/memory/update.py	Uses resolved user (including tool_context) to scope updates via owner_id where applicable.
src/adk_redis/tools/memory/delete.py	Wires tool_context into user resolution and updates documentation around user_id.
src/adk_redis/tools/memory/prompt.py	Uses resolved user (including tool_context) to scope prompt enrichment memory search.
docs/concepts/memory.md	Documents the updated user resolution order for tools running inside an ADK agent loop.

Suppressed comments (1)

src/adk_redis/tools/memory/delete.py:125

• tool_context, _get_namespace(...), and _get_user_id(...) are called but their results are unused. Since these helpers have no side effects, these lines are effectively dead code and suggest that namespace/user scoping is applied when it is not.

    tool_context = kwargs.get("tool_context")

    memory_ids = args.get("memory_ids", [])
    self._get_namespace(args.get("namespace"))
    self._get_user_id(args.get("user_id"), tool_context=tool_context)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

Suppressed comments (2)

src/adk_redis/tools/memory/prompt.py:132

• MemoryPromptTool now resolves the user via tool_context, but there are no tests covering this path (unlike create/search/update/delete). Adding a prompt-tool test that asserts the generated request is scoped to tool_context.user_id would prevent regressions in multi-user deployments.

    args = kwargs.get("args", kwargs)
    tool_context = kwargs.get("tool_context")

    query = args.get("query")
    system_prompt = args.get("system_prompt", "")
    namespace = self._get_namespace(args.get("namespace"))
    user_id = self._get_user_id(args.get("user_id"), tool_context=tool_context)

src/adk_redis/tools/memory/delete.py:140

• DeleteMemoryTool validates scope by fetching each memory one-by-one before deleting. For large memory_ids lists this adds N extra round-trips (and is sequential), which can significantly increase latency and load. Consider using a backend bulk lookup, or parallelizing the preflight checks with bounded concurrency if per-id validation is required.

      if self._config.backend == OPENSOURCE_AGENT_MEMORY_BACKEND:
        client = self._get_agent_memory_server_client()
        for memory_id in memory_ids:
          memory = await client.get_long_term_memory(memory_id=memory_id)
          self._require_memory_scope(
              memory,
              namespace=namespace,
              user_id=user_id,
          )
        response = await client.delete_long_term_memories(memory_ids=memory_ids)

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/adk_redis/tools/memory/delete.py:151

• memory_ids is assumed to be a list of non-empty strings, but tool args can arrive with the wrong type (for example a single string). In that case, the current code will treat the string as an iterable and issue backend calls for individual characters, leading to confusing failures or unintended behavior. Validate the input type and elements and return a clear error before proceeding.

    memory_ids = args.get("memory_ids", [])
    namespace = self._get_namespace(args.get("namespace"))
    user_id = self._get_user_id(args.get("user_id"), tool_context=tool_context)

    if not memory_ids: