fix: resolve empty response display and streaming errors

[jalen0x]

Summary

Fixes three related bugs that cause (No content to display) in Telegram responses when Claude performs tool-heavy operations.

Changes

1. Add helper methods to StreamUpdate dataclass (sdk_integration.py)

The message handler calls get_tool_names(), is_error(), get_error_message(), and get_progress_percentage() on StreamUpdate objects, but these methods did not exist. Added all four methods plus a progress attribute.

2. Fix ConversationEnhancer call signatures (message.py)

The caller was passing wrong arguments to update_context(), should_show_suggestions(), and generate_follow_up_suggestions(). Aligned all calls to match the actual method signatures in conversation_mode.py.

3. Generate tool summary for empty responses (sdk_integration.py)

When Claude's response is purely tool operations (file edits, bash commands) with no conversational text, content was empty. Now generates a fallback: ✅ Task completed. Tools used: Edit, Bash, ...

Testing

• python -m compileall -q src passes
• Only 2 files modified, 121 additions, 15 deletions

Fixes #135

[FridayOpenClawBot]

This fixes the three root causes from #135 correctly — the missing StreamUpdate methods, wrong ConversationEnhancer call signatures, and empty content when Claude only uses tools. The approach is sound.

Issues to address before merging:

StreamUpdate is a dataclass with instance methods — this works but is unusual for a data-transfer object. If StreamUpdate is used across multiple call sites or serialised/deserialised, the methods travel with it fine, but consider whether conversation_mode.py should own the logic for interpreting ClaudeResponse rather than StreamUpdate owning logic about itself. Not a blocker, just worth flagging.

is_error() false-positive risk — this check:

if self.metadata.get("error") or self.metadata.get("error_message"):
    return True

will return True if either key exists with any truthy value, including a dict {} or int 0-adjacent truthy values. If the SDK ever includes an "error": null or "error": {} in normal metadata, this won't trigger (falsy), but "error": "none" would. Low risk in practice but worth a type guard: isinstance(self.metadata.get("error"), str).

Tool summary fallback is user-facing with no i18n hook — "✅ Task completed. Tools used: {tools_summary}" is hardcoded English. Fine for now, but if localisation is ever added, this will be a find-and-replace problem. A constant would be cleaner.

get_or_create_context(user_id) return type — the diff shows conversation_context is now assigned from get_or_create_context() rather than the return value of update_context(). Verify that get_or_create_context() returns the same context type that generate_follow_up_suggestions() expects — if the type changed, mypy strict mode should catch it, but worth a manual check if CI isn't running mypy on this path.

Positive: The _to_int helper in get_progress_percentage() is well-written — handles int, float, and str gracefully with proper clamping. The deduplication via dict.fromkeys() is clean throughout. The str(result_content).strip() fix is minimal and correct.

Overall this is a legitimate bug fix. Address the is_error() type guard and verify the context type alignment, then it's good to go.

— Friday, AI assistant to @RichardAtCT

[FridayOpenClawBot]

PR Review
Reviewed head: de1550b236eec8d57bffcfa3d37a6618a056e8cf

Summary

• Added isinstance(str) type guards to is_error() for both error and error_message metadata fields
• Extracted hardcoded fallback string as TASK_COMPLETED_MSG module-level constant

What looks good

• Type guards correctly fix the false-positive risk — {}, 0, [] etc. no longer trigger is_error(). The two-variable split is clear and readable.
• TASK_COMPLETED_MSG placement at module level with a descriptive comment is clean. Using .format() at the call site keeps the constant itself a simple template string.
• Commit message explicitly calls out the type compatibility verification for get_or_create_context / generate_follow_up_suggestions — good discipline even if no code change was needed.

Issues / questions
None. Both pieces of feedback addressed correctly.

Verdict

• ✅ Ready to merge

— Friday, AI assistant to @RichardAtCT