# AI use cases for Tracecrate (FankChen/tracecrate)

Repo: [FankChen/tracecrate](https://github.com/FankChen/tracecrate) | Use cases: 5 | Page: https://dedupedai.com/use-cases/fankchen-tracecrate/

TraceCrate is a local-first, client-side workbench that reads AI agent trace files and shows a searchable timeline, recorded metrics, heuristic diagnostics, and side-by-side comparisons.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Read a Claude Code or Codex session step by step: https://dedupedai.com/use-cases/fankchen-tracecrate/inspect-own-agent-trace-locally/
- Compare two agent runs event by event: https://dedupedai.com/use-cases/fankchen-tracecrate/compare-two-agent-runs/
- Find failing and slow tool calls in an agent run: https://dedupedai.com/use-cases/fankchen-tracecrate/find-failing-and-slow-tool-calls/
- Export a trace report with structure only: https://dedupedai.com/use-cases/fankchen-tracecrate/share-trace-report-with-privacy-boundary/
- Explore synthetic agent runs without installing anything: https://dedupedai.com/use-cases/fankchen-tracecrate/explore-demo-traces-in-browser/

## How to read a Claude Code or Codex session step by step with Tracecrate

Repo: [FankChen/tracecrate](https://github.com/FankChen/tracecrate) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Load one authorized agent session file and browse its events in a local timeline.

**Who it is for:** A developer who ran Claude Code or Codex on a task and wants to see the exact sequence of messages, tool calls, and results.

**The problem:** Raw agent transcripts are long JSONL files that are hard to scan for what the agent actually did.

**The outcome:** A searchable, filterable timeline of the session where tool inputs, outputs, and recorded errors can be opened and read.

### Steps

1. Install Node.js 22.12 or newer (24 recommended) and npm.
2. Run `git clone https://github.com/FankChen/tracecrate.git`
3. Run `cd tracecrate`
4. Run `npm ci`
5. Run `npm run dev`, then open the address Vite prints.
6. Click Choose files, select only a session file you are authorized to inspect, then open a tool event in Timeline.

### Prompt to paste into your agent

```text
Using the TraceCrate repo at <repo path>, run npm ci and npm run dev, then load the authorized session file <session file path> through Choose files. Expected output: a Timeline listing that session's events with search and kind filters working. Verify that token totals appear only where the file reports them and that missing values are shown as unknown, not zero.
```

### Caveat

Claude session files are private, so only select files you are authorized to inspect, and TraceCrate does not scan agent directories for you.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/fankchen-tracecrate/inspect-own-agent-trace-locally/

## How to compare two agent runs event by event with Tracecrate

Repo: [FankChen/tracecrate](https://github.com/FankChen/tracecrate) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Load two sessions and see recorded metric differences and aligned tool calls.

**Who it is for:** An engineer testing a prompt, configuration, or code change who wants to know how two agent runs differ.

**The problem:** Comparing two long agent runs by hand means matching events and metrics across two separate transcripts.

**The outcome:** A comparison view showing recorded metric differences, tool-call counts, and changed fields for aligned events, with links back to the original events.

### Steps

1. Run `npm ci` and `npm run dev` in the cloned Tracecrate folder, then open the printed address.
2. Click Choose files and select the baseline session and the changed session together, up to 5 files per selection.
3. Open Insights, then Compare.
4. Scroll to Event-by-event comparison and check changed fields, or choose Tools only.
5. Click an event button to jump to the original session and timeline page.

### Prompt to paste into your agent

```text
Using the TraceCrate repo at <repo path>, start the dev server and load the two authorized session files <baseline file> and <changed file>. Expected output: the Compare view with recorded metric differences shown as B minus A and an event-by-event list. Verify that the comparison is described as descriptive, since repeated tool names can align ambiguously and this is not a controlled benchmark.
```

### Caveat

The README labels its demo baseline and optimized runs as manufactured, so comparisons show what was recorded, not measured improvements.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/fankchen-tracecrate/compare-two-agent-runs/

## How to find failing and slow tool calls in an agent run with Tracecrate

Repo: [FankChen/tracecrate](https://github.com/FankChen/tracecrate) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Filter a session for reported errors and long durations, then review heuristic patterns.

**Who it is for:** A developer whose agent run took too long or stopped partway and who wants to locate the costly steps quickly.

**The problem:** Slow or failing tool calls are buried among hundreds of routine events.

**The outcome:** A short list of the longest tool events and any explicit tool errors, plus heuristic flags for repeated or large outputs.

### Steps

1. Run `npm run dev` in the Tracecrate folder and open the printed address.
2. Load the session file through Choose files.
3. In Timeline, select Reported error or set Min duration (ms), then choose Longest first.
4. Open Insights to review explicit tool errors, long tool durations, and large outputs.
5. Open each flagged event to read its recorded inputs and outputs before drawing conclusions.

### Prompt to paste into your agent

```text
Using the TraceCrate repo at <repo path>, start the dev server and load <session file path>. Expected output: a Timeline filtered to reported errors or events above a chosen minimum duration, sorted longest first. Verify that events with missing duration are excluded when a minimum is set, and that heuristic flags are treated as hints, not confirmed causes.
```

### Caveat

Heuristics do not establish causes or intent, so a flagged slow tool call still needs human reading of the original event.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/fankchen-tracecrate/find-failing-and-slow-tool-calls/

## How to export a trace report with structure only with Tracecrate

Repo: [FankChen/tracecrate](https://github.com/FankChen/tracecrate) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Create a shareable JSON or HTML report that removes free text and original identifiers.

**Who it is for:** A team member who must share an agent trace in a bug report or review but cannot send the raw transcript.

**The problem:** Raw traces contain prompts, outputs, and identifiers that should not leave the machine by default.

**The outcome:** A structure-only JSON or standalone HTML report with a previewed boundary, ready to share after a full review of the download.

### Steps

1. Run `npm run dev` and load the trace through Choose files.
2. Open Export report and select Structure only.
3. Optionally select Omit timing metadata and/or Omit token usage.
4. Click Show redacted preview, then download JSON or a standalone HTML report.
5. Open the full downloaded file and inspect every field before sharing it.

### Prompt to paste into your agent

```text
Using the TraceCrate repo at <repo path>, load <session file path>, export a Structure only report with timing and token minimization enabled, and save it as <output path>. Expected output: a JSON or HTML file with no raw free text. Verify by reading the full download, since event order, counts, and statuses remain and can still be sensitive.
```

### Caveat

Neither export mode guarantees anonymization, and switching to patterns clears the timing and token minimization.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/fankchen-tracecrate/share-trace-report-with-privacy-boundary/

## How to explore synthetic agent runs without installing anything with Tracecrate

Repo: [FankChen/tracecrate](https://github.com/FankChen/tracecrate) | Difficulty: Beginner | Time: 15 minutes | Basis: From the README

Open the hosted demo and follow the 30-second walkthrough on synthetic runs.

**Who it is for:** A curious reader or evaluator who wants to see what a trace workbench does before deciding to install it.

**The problem:** It is hard to judge a trace tool from its description alone.

**The outcome:** A quick tour of Timeline, Insights, Compare, and export using the two synthetic pagination runs that load automatically.

### Steps

1. Open https://fankchen.github.io/tracecrate/ in a desktop browser.
2. In Timeline, open a tool event and try a search or a kind filter.
3. Open Insights, then Compare, and read the recorded differences.
4. Choose Export report, then Structure only, then Show redacted preview.
5. Remember that the demo labels baseline and optimized are manufactured, not measured improvements.

### Prompt to paste into your agent

```text
Open the hosted TraceCrate demo at https://fankchen.github.io/tracecrate/ and follow its walkthrough on the two synthetic pagination runs. Expected output: a short summary of the Timeline filters, the Compare differences, and the redacted export preview. Verify that the summary does not describe the demo labels as measured results.
```

### Caveat

The hosted demo uses only synthetic data, and a static host still receives ordinary request metadata such as IP address and user agent.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/fankchen-tracecrate/explore-demo-traces-in-browser/