fix(pipecat): stop re-storing injected memory blocks as new memories

[abhay-codes07]

What

The pipecat service was writing its own injected memories back to Supermemory as brand-new memories.

LLMContext.get_messages() returns the context's live message list — the enhancement code depends on that, mutating the returned list in place to inject the <user_memories> block. But process_frame captured that same reference before enhancement:

context_messages = context.get_messages()      # live reference
...
self._enhance_context_with_memories(...)       # appends the memory block
...
storable_messages = [msg for msg in context_messages if msg["role"] in ("user", "assistant")]

By the time storable_messages was sliced, the list already contained this turn's injected user-message, so _store_messages persisted the recalled memories as a new memory. The next turn recalls the stored block and re-stores it again — a self-referential loop that compounds every turn. (It was previously masked by the dedup crash fixed in #1268; with profile facts present and search results empty, the loop is live.)

Fix

Two defenses, both needed:

• snapshot — list(context.get_messages()) before enhancement, so the current turn's injection never enters the storage slice
• filter — a _is_memory_injection helper excludes any message carrying the memory tag from storable_messages, which covers the previous turn's injected block still sitting in the context (it's only replaced during the next enhancement, after the snapshot is taken)

The helper is tolerant of non-string content (multimodal parts, None).

Testing

• New tests/test_no_memory_restorage.py drives process_frame end-to-end with a live-list fake context (same semantics the enhancement code relies on) and a mocked Supermemory client: the memory block lands in the context but never in the storage payload, for both the same-turn and next-turn shapes; plus non-string-content tolerance for the helper
• python -m pytest tests in the package — 4 passed
• No conflict with fix(python-sdks): stop mishandling SDK model objects in memory dedup and formatting #1268 (that PR touches utils.py; this touches service.py)

cc @MaheshtheDev

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.