Improve workspace experience in OpenClaw

[andyrewlee]

Summary

Describe the change and intended behavior.

Quality Checklist

• Ran make devcheck locally.
• Ran make lint-strict-new locally for changed code.
• If UI/rendering changed, ran make harness-presets.
• If tmux/e2e changed, ran go test ./internal/tmux ./internal/e2e.

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 5 additional findings.

[devin-ai-integration]

Devin Review found 1 new potential issue.

View 18 additional findings in Devin Review.

[devin-ai-integration]

🔴 shouldDropAgentChromeLine drops all lines starting with "> " which silently strips legitimate agent output

The new strings.HasPrefix(clean, "> ") check in shouldDropAgentChromeLine causes any line starting with "> " (greater-than + space) to be treated as chrome noise and silently dropped from agent output. This prefix is extremely common in legitimate content: Markdown blockquotes (> quoted text), shell prompt echoes (> cd /tmp), diff output context lines, and general agent prose that starts with >. The function is called from compactAgentOutput, detectNeedsInput, detectNeedsInputPrompt, trailingNeedsInputOptionLines, and lastNonEmptyLine — meaning it affects summaries, needs-input detection, delta computation, and the latest-line field in wait responses.

Root Cause and Impact

The intent was to filter droid TUI prompt lines like > Reply exactly READY in one line. (visible in the test at cmd_agent_output_signals_test.go:46). However, "> " is a standard Markdown blockquote prefix and appears in many contexts:

• Agent output quoting user instructions: > The user asked to fix the login bug
• Shell command echoes: > npm run build
• Diff context lines: > +func newHelper()
• Conversational quoting: > As mentioned earlier...

All such lines will be silently dropped from compactAgentOutput() results. Since shouldDropAgentChromeLine is used in lastNonEmptyLine (cmd_agent_output_signals.go:383), detectNeedsInput (cmd_agent_output_signals.go:189), and the wait-response builder (cmd_agent_wait_response.go:231), this causes:

1. Summary/latest_line fields may be empty or skip important content when the most recent meaningful output starts with >.
2. needs_input detection may miss prompts if the input hint line begins with >.
3. Delta output loses blockquoted sections, making chat summaries incomplete.

Impact: Agent output is silently truncated for any assistant that uses Markdown blockquotes or shell echo lines, leading to incomplete summaries sent to chat orchestrators.

Prompt for agents

In internal/cli/cmd_agent_output_signals.go line 106, the check `strings.HasPrefix(clean, "> ")` is too broad. It should be narrowed to only match the specific droid TUI prompt pattern. Options:

1. Remove the `"> "` prefix check entirely and instead add a more specific pattern like checking for droid prompt echoes (e.g., lines that look like `> <user prompt text>` preceded by droid chrome).

2. Or, make the check more specific by requiring additional context, such as:
   - `strings.HasPrefix(clean, "> ") && len(clean) > 80` (only long prompt echoes)
   - Or keep it but only inside a droid-specific context flag

The safest fix is to remove line 106 (`strings.HasPrefix(clean, "> "),`) since the droid prompt line `> Reply exactly READY...` from the test would still be handled by the existing test's expected output (the test shows the `>` line IS dropped, but the actual READY output lines are preserved). The test at line 46 of cmd_agent_output_signals_test.go should be updated to verify the `> Reply...` line is NOT dropped (or the filter should use a narrower pattern).

---

Was this helpful? React with 👍 or 👎 to provide feedback.