From epic to conversation replay

Applies to: Dispatch v1.0.0, last verified 2026-04-05

This guide shows you how to navigate from an epic card to the full conversation replay for any phase, including the messages, tasks, file changes, and agent spawn tree.

TL;DR: Click an epic card to open the drawer. Find the linked session in the phase detail. Click the session to open the session drawer. Read the full conversation, task list, and file changes there.

What you are looking at

Dispatch links sessions to phases automatically. When an agent (from any supported provider) works on a phase, it runs as a session. Dispatch matches that session to the phase by correlating timestamps in the execution log with session activity windows. The result is a linkedSessions list on each phase, which Dispatch uses to navigate you directly to the conversation.

The path has four stops:

Epic card → Epic drawer → Phase entry → Session drawer → Conversation

Step 1: Open the epic drawer

Click any epic card in the center column. The Epic Drawer slides in from the right.

The drawer shows:

The full plan prose (the body of plan.md, rendered as Markdown)
The parsed execution log, one entry per completed phase

Step 2: Find the phase you want to review

Scroll through the execution log panel in the drawer. Each completed phase has a header like:

[2026-04-03T14:32:00Z] Phase 2: Implement the API — @staff-engineer

Below the header is the summary the agent wrote: what it did, what it found, and what the next phase needs to know.

If the phase has linked sessions, a Sessions row appears below the summary showing the session IDs that Dispatch matched to that phase window.

Two things prevent a session from appearing here:

Wrong branch name. The most common cause. Dispatch links sessions to phases partly by checking whether the session’s git branch contains the epic slug (the kebab-case name of the .tasks/<epic-name>/ directory). If the branch you ran the agent on doesn’t include that slug, the session won’t link. Check the cyan branch badge on the session card and compare it to the epic name. Switching to or running on a branch like feat/<epic-slug> fixes this on the next refresh.
Session data is gone. Session linking requires the session to still exist in the provider’s local history (e.g., Claude Code’s JSONL files, Gemini chat JSON, Codex logs). If the session is very old or the data was deleted, the link won’t appear.

The execution log entry remains in full regardless of whether sessions are linked.

Step 3: Open the session drawer

Click any linked session ID or session badge. The Session Drawer replaces the epic drawer (or opens alongside it, depending on viewport width).

The session drawer shows:

Section	Content
Header	Full opening prompt, activity status, project name, git branch, session ID
Details	Working directory path, project name, full session ID
Tasks	Every task the agent created with `TaskCreate`, with status icons
Subagents	Any child agents this session spawned, each expandable
Agent spawn graph	Visual tree of parent → child agent relationships (click the graph icon)

Step 4: Read the conversation

Scroll down past the header and details. The Messages section shows the conversation excerpt: alternating user and assistant turns, capped at the most recent entries.

Each turn shows:

The role (you / agent name)
The message text, with long messages truncated and expandable

This is the actual conversation that produced the phase output. You can see the prompts the agent received, the reasoning it wrote, the tool calls it made, and the responses it got back.

Step 5: Review file changes (optional)

If the agent used Write or Edit tools during the session, the Files Modified section lists every file it touched, with line change counts.

Click any file row to expand the diff view. The diff shows added lines in green and removed lines in red, grouped by hunk.

This is particularly useful after a large refactoring phase: you can see exactly what the agent changed in each file without leaving the browser.

Reading the agent spawn graph

If the session spawned subagents (using the Agent tool), a small graph icon appears in the session drawer header.

Click the icon to open the Agent Spawn Tree: a modal with the parent session as the root and each subagent as a child node. Edges are labelled with the description the orchestrator provided when spawning the agent.

Edge style	Meaning
Solid line	Subagent was launched synchronously (orchestrator waited for it to finish)
Dashed line	Subagent was launched in the background (fire-and-forget)

Active nodes glow; idle nodes are dimmed. Click any subagent node to open that subagent’s own session view.

Navigating across sessions for the same phase

A phase can have multiple linked sessions when an agent was restarted mid-phase. This happens, for example, after a context recovery as described in Recovering abandoned work.

In this case the phase shows multiple session badges. Each one is a separate conversation. Review them in chronological order to follow the full arc: first session starts the work, later sessions continue from where the previous one left off.

Getting to a session without going through an epic

Two other entry points reach the session drawer directly:

From the Sessions tab: Click the Sessions tab in the center column. Every live and recent session appears as a card. Click any card to open its drawer. This gives you session details without needing to know which epic the session was part of.

From the agent sidebar: Click a persona in the left sidebar to open the Agent Profile Panel on the right. The Live Sessions section lists active sessions for that persona. Click any entry to open the session drawer.

Key	Action
Escape	Close the current drawer (session drawer first, then epic drawer)
Escape (second press)	Close the spawn graph if open, then close the drawer on the next press
Enter or Space on a focused card	Open the session drawer for that card
Click backdrop	Close the current drawer