All Docs
FeaturesSaaS FactoryUpdated March 11, 2026

Agent Activity Transparency — Diff Viewer, Reasoning Chain & Files Changed

Agent Activity Transparency

Version 1.0.126 introduces the Agent Activity Transparency feature set — a Cursor/Devin-style visibility layer that shows exactly what every AI agent is doing, why it made each decision, and what code it changed.

All new UI builds on top of the existing SSE-streamed observability data; there are no backend schema changes or new API routes.

Overview

Previously, the pipeline detail view exposed a flat "Activity Feed" of tool calls. As of v1.0.126 the feed is reorganised into a 3-tab "Agent Explainability" panel and the agent job detail dialog gains two additional tabs, giving enterprise approvers a complete audit trail:

QuestionWhere to find it
WHY did the agent choose this feature?Reasoning tab → WHY phase
WHAT code did it change?Files Changed tab → syntax-highlighted diffs
HOW did it make each decision?Reasoning tab → decision chain
WHEN did each action happen?Timestamps on every step

New Components

DiffViewer

A syntax-highlighted, LCS-based code diff viewer.

Key capabilities:

  • Displays + (added) and − (removed) line counts in a header badge.
  • Green/red line-level highlighting for additions and deletions.
  • Collapsible unchanged-context lines to reduce visual noise.
  • One-click copy button for the full file content.
  • Automatic language detection for 20+ file types, including TypeScript, JavaScript, Python, JSON, SQL, Markdown, YAML, and more.
  • DiffBadge — a compact inline component that renders a file path with language indicator, used inside tool-call log rows.

ReasoningChain

A timeline component that visualises the agent's internal reasoning as a WHY → WHAT → HOW progression.

Key capabilities:

  • Icon bubble + connector timeline layout.
  • Per-level styling:
    • Thinking — purple
    • Decisions — blue
    • Tool calls — neutral
    • Insights — amber
    • Errors — red
  • Long entries are collapsible to keep the view scannable.
  • "Show more" pagination for jobs with many reasoning steps.

Live Pipeline Detail — Agent Explainability Panel

The panel previously labelled Activity Feed is now Agent Explainability and splits into three tabs.

Activity Tab

Structured tool-call rows with smart rendering per tool type:

  • write_file — renders an inline DiffViewer showing exactly what was written.
  • search_code — shows the search query as a code preview.
  • create_pull_request — displays the PR title.
  • All other tools fall back to a compact row with tool name and key parameters.

Reasoning Tab

The full ReasoningChain for the selected pipeline run, grouped into WHY / WHAT / HOW phases with timestamps.

Files Changed Tab

An aggregated list of every write_file call across the entire pipeline run. Each file is expanded into a DiffViewer so reviewers can inspect every line change without opening individual job dialogs.

Running Job Cards

  • A live "thinking" bubble appears on cards for jobs that are currently running, streaming the agent's latest reasoning step in real-time.
  • A "X files written" badge shows a running count of file writes on each job card.

Agent Job Detail Dialog

The detail dialog (opened by clicking any agent job row) gains two new tabs alongside the existing Trace tab.

Files Changed Tab

Lists every write_file tool call recorded for that job. Each entry renders a DiffViewer with syntax highlighting instead of the previous raw JSON dump, making it practical to review code changes directly in the dialog.

Reasoning Tab

Shows the agent's decision chain — the ordered sequence of meaningful actions (file reads, file writes, PR creation, etc.) — with:

  • A phase legend distinguishing reading vs. creating phases via coloured indicators.
  • Timestamps for each action.
  • Collapsible entries for long reasoning text.

Trace Tab (updated)

write_file tool-call entries in the existing Trace tab now render a DiffViewer instead of raw JSON input, consistent with the rest of the UI.

Technical Notes

  • No backend changes — all new UI consumes data already present in the SSE stream.
  • No database schema changes.
  • TypeScript compiles cleanly with this release.
  • The safeStr() helper was added to live-pipeline-detail.tsx to prevent unknown values from being passed directly as React children.