feat(dashboard): surface sandbox stdout/stderr in task timeline

[Abernaughty]

Summary

Repurposed from "terminal agent blades" — the per-agent SSE-filtered terminal tabs are redundant with the existing task timeline and agent sidebar. What's actually missing is sandbox output visibility: when the sandbox runs a script, the dashboard shows "Sandbox: 1 tests passed" but never displays what the script actually printed.

This was identified during trace analysis of the triforce task where QA failed the output for missing leading spaces — a problem that would have been immediately obvious if sandbox stdout were visible in the timeline.

Current State

• SandboxResult.output_summary captures stdout (+ stderr on failure), already secret-scanned and truncated at 2000 chars
• SandboxResult.errors contains extracted error messages (tracebacks, SyntaxError, etc.)
• runner.py handles sandbox_validate but does not include output_summary in the SSE payload — only pass/fail counts
• The data exists end-to-end in GraphState, it just isn't rendered

Implementation

Backend (runner.py) — Tiny

In the sandbox_validate SSE event handler, include output content:

sandbox_result = output.get("sandbox_result")
if sandbox_result:
    event_data["output_summary"] = sandbox_result.output_summary
    event_data["errors"] = sandbox_result.errors
    event_data["exit_code"] = sandbox_result.exit_code

Dashboard (Svelte) — Small-Medium

1. Collapsible <pre> block beneath sandbox timeline events when output_summary is non-empty
2. Auto-expand on failure (exit_code !== 0) — user needs to see what went wrong without extra clicks
3. Monospace rendering — DM Mono, dark background (#0a0c10), matching terminal panel styling
4. Truncation — max 50 lines with "Show all (N lines)" toggle
5. Error display — errors list rendered above stdout in red-tinted block (#ef4444 bg) when present
6. Copy button — top-right of the output block
7. UI-side redaction — regex pass matching secret patterns before rendering (Layer 4 of defense model, belt-and-suspenders with server-side scanning)
8. Empty state — if output_summary is empty/whitespace, render nothing (no empty collapsible)

Files to Modify

File	Change	Effort
dev-suite/src/api/runner.py	Add output_summary, errors, exit_code to sandbox SSE event	Tiny
dashboard/src/lib/stores/tasks.svelte.ts	Store sandbox output fields from SSE	Tiny
dashboard/src/lib/components/ (timeline event component)	Collapsible output block, auto-expand, copy button	Small-Medium

Test Plan

• Unit: Mock SSE event with output_summary → verify store update
• Manual: Run triforce task → confirm ASCII art visible in collapsed block under sandbox event
• Manual: Run a task that crashes → confirm stderr/traceback auto-expands in red
• Manual: Run a task with >50 lines output → confirm truncation + "Show all"
• Edge: Empty stdout script → confirm no empty collapsible renders

Not in Scope

• Separate stderr field on SandboxResult — not needed, already combined in output_summary
• QA verdict / exact_fix_hint display alongside sandbox output — separate enhancement
• Terminal resize bug (bug: Terminal panel drag-to-resize breaks after ~1/3 screen height #113) and real shell (feat: Terminal real shell — WebSocket PTY, multiple instances, localhost-only #114) — unchanged, still on roadmap

Effort

Small-Medium (mostly dashboard rendering, trivial backend addition)

Dependencies

None