More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Session Replay for Claude Code

    1

    Turn any Claude Code session into an offline HTML replay: every prompt, subagent and swarm on a map and a timeline, with start and finish times and token counts.

    Free

    Security scanned
    Session Replay for Claude Code

    Session Replay for Claude Code

    Example session with this skill installed

    Replay the session where we ran the sorting-algorithm swarm. I want a timeline of when each agent started and finished, and how many tokens they used.

    • Read your context and instructions
    • Compiled the session replay
    • Generated the document

    Here's the replay of your swarm session, "Set up a benchmark harness: random, sorted, reversed and nearly sorte…".

    2 prompts, 11 subagents, 75 tool calls (1 failed), 2.0M tokens in, 76k out.

    AgentStartedFinishedTookTool callsTokens in / out
    Main agent, prompt 111:55:5711:57:101m 13s6185k / 7k
    Main agent, prompt 212:13:57still running50m 12s12– (counted when it stops)
    E1 Quicksort12:14:4412:44:4430m 00s8270k / 8k
    E2 Merge sort12:14:5112:52:5138m 00s8290k / 15k
    E3 Timsort12:14:5812:38:5824m 00s8178k / 6k
    E4 Radix sort12:15:0512:59:0544m 00s9294k / 14k
    C3 Measurer E03_timsort12:39:2312:45:185m 55s3122k / 3k
    C3 Measurer E01_quicksort12:45:0912:50:145m 05s386k / 4k
    C4 Fact-check E03_timsort12:45:3812:50:094m 31s4176k / 4k
    C4 Fact-check E01_quicksort12:50:3412:56:135m 39s4204k / 3k
    C3 Measurer E02_merge12:53:1612:57:504m 34s393k / 6k
    C4 Fact-check E02_merge12:58:1013:04:096m 00s4111k / 6k
    C3 Measurer E04_radix12:59:30still running4m 39s3– (counted when it stops)

    What stands out

    • The four explorers ran in parallel, all starting within 21 seconds (12:14:44 to 12:15:05).
    • E4 Radix sort was the slowest and the hungriest: 44 minutes and 294k tokens in. Its measurer is still running.
    • Each explorer handed over to a measurer, then a fact-checker, within a minute of finishing; the timsort chain finished first.
    • 1 tool call failed; the replay marks it in red on the timeline.

    Your replay: /home/sam/code/algo-lab/session-replay-2026-11-09-demo-sor.html

    One file, opens offline in any browser:

    Map: who started whom, with the main agent in blue and every subagent in orange. Press Play to watch the swarm unfold.

    The map: four explorers, each handing over to a measurer and a fact-checker

    • Timeline: one bar per agent from start to finish, with a tick for each tool call and red for failures.

    The timeline: one bar per agent with durations, tool-call ticks and token counts

    • Story: each agent's task, tools, tokens and report, in order.

    Next steps

    • Want me to open it in your browser?
    • I can replay another session too; say which one, or I'll list them.
    • If you share the file, read it first: it contains your prompts and the agents' reports.

    session-replay-2026-11-09-demo-sor.html

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Ever wondered what actually happened while Claude Code was working? One prompt can start five subagents, a code reviewer and a whole workflow swarm, and the chat only shows you the summary.

    Session Replay turns a Claude Code session into a single HTML file you can open in any browser, replay like a video, and send to a colleague. Just ask: "replay my last session" or "show me a timeline of what the agents did".

    What you see

    • Every prompt you typed, plus the questions Claude asked back and what you answered.
    • The map: the main agent in blue, and every subagent and workflow agent it started in orange, each linked to whoever started it. Swarm agents show their own names and phases.
    • The timeline: one bar per agent, from when it started to when it finished, with a tick for every tool call and red marks where something failed. Like a Gantt chart for your agents.
    • Times and tokens: start and finish times, durations, tool calls and token counts (input, cache and output) per agent and per prompt.
    • Replay controls: play, pause and scrub through the session as it happened. Click a prompt to see only the agents it set off.

    How it works

    The skill runs a small Node.js script (no packages) that reads the chosen session's transcript, which Claude Code already keeps on your computer, and writes one HTML file. Use --list to pick an older session by its first prompt.

    Private by design

    • Reads only the selected project's own Claude Code transcripts. Nothing else.
    • No network calls, nothing installed, no settings changed.
    • Keeps only tool names and one-line summaries, never file contents or command output.
    • Anything that looks like a secret (tokens, passwords, API keys, URL query strings) is masked as ***.

    Requirements: Claude Code and Node.js 18 or newer.

    Want it live? This skill replays a session on request. For a page that follows all your sessions in real time, with alerts when one needs your input, get the free Agent Monitor plugin: https://github.com/omer-parvez/claude-agent-monitor

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    Requires Node.js 18+. Limited to Claude Code project transcripts. Masks secrets and truncates text to 300 characters.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 7 days ago

    • Free with an Agensi account

    Software
    Claude CodeNode.js 18+

    Reads only the chosen project's Claude Code transcripts (~/.claude/projects/) and writes one HTML file. No network, no packages. Run it from the project folder you used Claude Code in, or pass --project folder . The file includes your prompts.

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Environment Variables

    Allowed Hosts

    www.w3.org
    Listed7 days ago

    What's inside

    Frequently Asked Questions

    This skill solves the 'black box' problem of complex agent sessions by converting text logs into a visual timeline and map, making it easy to identify where subagents failed, looped, or consumed excessive tokens.

    No, the output is a single, standalone HTML file that runs entirely offline with no external dependencies, ensuring your logs and session data never leave your local environment.

    It is specifically optimized for Claude Code transcripts; it requires Node.js 18+ to parse the local project directory and generate the interactive visualization.

    The tool includes a built-in privacy scrubber that automatically masks API keys, secrets, and bearer tokens, while truncating long prompts to ensure sensitive data isn't exposed in the visual report.

    You receive the full Node.js parsing script, the visualization engine, and documentation on how to integrate the generator into your local development workflow.

    New in Agents & Orchestration

    Free