Summary

Fixes #6316 by keeping built-in memory tool lifecycle previews out of public gateway progress surfaces while preserving normal tool execution and private model-visible tool results.

Root Cause

The memory tool already generated concise previews such as +memory, ~memory, and -memory. Those previews are useful internally, but gateway progress renderers treated them like ordinary tool progress and forwarded them to public-facing chat surfaces, including Telegram and API streaming clients.

Changes

• Add a shared should_hide_public_tool_progress() policy in agent.display.
• Suppress memory progress before Telegram/gateway progress messages are rendered, including verbose argument rendering and long-tool onboarding hints.
• Suppress memory lifecycle events in OpenAI-compatible chat-completions SSE progress.
• Apply the same privacy rule to Responses API function-call streaming and /v1/runs structured progress events.
• Add regression tests covering Telegram all/verbose modes, memory completion-only hints, chat SSE, Responses SSE, and run progress callbacks.

Validation

• uv run --extra dev --extra messaging pytest tests/agent/test_display.py tests/gateway/test_memory_tool_progress_privacy.py tests/gateway/test_api_server.py::TestChatCompletionsEndpoint::test_stream_tool_progress_skips_memory_events tests/gateway/test_api_server.py::TestResponsesStreaming::test_stream_function_call_items_skip_memory_tool tests/gateway/test_api_server.py::TestRunStructuredEvents::test_run_event_callback_skips_memory_tool_progress -q
• uv run --extra dev ruff check agent/display.py gateway/run.py gateway/platforms/api_server.py tests/agent/test_display.py tests/gateway/test_api_server.py tests/gateway/test_memory_tool_progress_privacy.py
• git diff --check HEAD