LOOPScope v0.1.0 — agent loop trajectory visualizer

LOOPScope v0.1.0

The first public release of LOOPScope — an agent-loop trajectory visualizer for
DeepSeek Harness (DSH).

Highlights

• LOOP view — one card per agent-loop iteration with an interactive four-phase
  ring (context → decision → act → observe), conversation/context/decision/act/
  observe chips, tool chips, developer-signal badges, a sticky LOOP sequence ribbon
  and a ▶ Play walkthrough.
• Duration bars — P90-normalized per-row durations so 100+ loop sessions stay
  readable in a narrow sidebar.
• Timeline — real wall-clock waterfall with idle compression, zoom and
  follow-latest.
• Developer details — route and header reason, context-window usage, tool
  roster, first-token latency, tok/s, retries/failed attempts, tool arguments,
  results and errors.
• Developer signals — approvals, workspace changes, deliverables, team tasks
  and subagents attached to the loop that produced them.
• zh/en locales and light/dark theme support; no runtime dependencies beyond React
  platform modules; no Host RPC and no network access.

Install

dsh plugin --profile web add github:hanranCode/dsh-loopscope
dsh --profile web

or via the DSH plugin manager (install_bundle target github:hanranCode/dsh-loopscope).
Open any session and click “轨迹图表” in the session header, then switch the
toolbar to 循环.

Verification

• Offline checks: npm test (artifact boundary + parser replay + zh/en server
  render of both views), CI on Node 20/22.
• End-to-end before release: validated in a separate dsh web instance against a
  live session — plugin loaded from the boot graph, 48/48 checks passed, LOOP view
  rendered with loops/rings/chips, inspectors and playback verified, zero console
  or page errors attributable to the plugin.

Notes

• Requires DSH >=0.2.0 <0.3.0.
• The plugin reads only the current session's durable event window; no file
  access, no network requests, no storage writes.

Full changelog: CHANGELOG.md