feat(signal): add reactions, native audio, and threading

[ishaanrathodd]

Summary

• add Signal-native reply context parsing plus opt-in native replies instead of forcing replies on every message
• expose react_message to Signal agents and route Signal reactions through sendReaction
• send Signal voice replies through stock signal-cli using filename-less data URI audio so they render in native audio UI
• fix Signal toolset expansion so react_message and text_to_speech are actually available at runtime
• update Signal prompting/session context so the agent can see replied-to content and choose human-like reactions

Details

This PR keeps Signal on the stock signal-cli path and improves three core interaction patterns:

1. Reactions

• adds a model-facing react_message tool
• defaults reactions to the current inbound message using session message metadata
• passes arbitrary emoji through Signal sendReaction

2. Replies / threading

• preserves inbound Signal timestamps as stable message_ids
• parses Signal quote metadata into reply context
• stops auto-threading Signal responses by default
• adds explicit opt-in native replies via [[reply_to_current]]

3. Voice / native audio UI

• sends Signal voice/audio replies as filename-less data URI attachments
• preserves stock signal-cli compatibility while rendering in Signal's native audio UI

Validation

• venv/bin/python -m pytest tests/hermes_cli/test_tools_config.py tests/gateway/test_base_topic_sessions.py tests/agent/test_prompt_builder.py tests/gateway/test_signal.py tests/test_toolsets.py -q
• venv/bin/python -m pytest tests/tools/test_send_message_tool.py -q

Both passed locally.

[britrik]

Code Review: PR #4768

Summary

Great work on adding Signal reactions, native audio, and threading support! The feature is well-designed and tests pass.

Positive Aspects

• Clean platform abstraction with get_default_reply_target() and requires_reply_context_metadata()
• Good test coverage with dedicated test classes for the new features
• Smart use of data URI encoding for native Signal audio UI rendering
• Proper environment variable injection for session context (message_id, user_id)

Potential Issues to Address

1. Memory Concern (signal.py): The _file_to_data_uri function reads entire files into memory with Path(file_path).read_bytes(). For large audio files, this could cause memory pressure. Consider streaming or adding file size limits.

2. Silent Error Handling (signal.py): The _coerce_timestamp function silently returns None when timestamp conversion fails. Consider logging a warning for debugging purposes.

3. Platform Limitation: The react_message tool only supports Signal currently - other platforms return an error. Consider documenting this limitation or implementing for other platforms (Telegram supports reactions via Bot API).

4. Test Assertion (test_signal.py): The test expects "voice.m4a" not in call["params"]["attachments"][0] - this is a good defensive check, but verify the data URI format works reliably with signal-cli across different audio formats.

Suggestions

• Add file size validation before _file_to_data_uri to prevent OOM
• Consider adding rate limiting for reactions to prevent spam
• Document the [[reply_to_current]] directive behavior in a user-facing format

Overall: LGTM with minor suggestions for improvement.

[ishaanrathodd]

britrik

Thanks — addressed the suggestions.

I added a safe fallback for oversized inline Signal audio, warning logs for invalid quote timestamps, and restricted react_message to active Signal sessions so the tool availability matches implementation scope.

Re-tested:
/hermes-agent/venv/bin/python -m pytest tests/gateway/test_signal.py tests/tools/test_send_message_tool.py tests/agent/test_prompt_builder.py tests/test_toolsets.py tests/hermes_cli/test_tools_config.py -q`

222 passed, 1 skipped

[ishaanrathodd]

Addressed a final Signal UX bug I found while testing:

• Signal auto-TTS voice replies no longer send the same response as text afterward
• implemented this as a platform hook so Signal gets voice-only behavior for auto voice replies, while other platforms keep their existing text+audio behavior
• added a regression test covering the Signal voice-input case

Re-ran the focused suite:
/.hermes/hermes-agent/venv/bin/python -m pytest tests/gateway/test_base_topic_sessions.py tests/gateway/test_signal.py tests/tools/test_send_message_tool.py tests/agent/test_prompt_builder.py tests/test_toolsets.py tests/hermes_cli/test_tools_config.py -q

Result: 231 passed, 1 skipped