Skip to content
Back to skills

Asciinema Cast Format

ASecurity

Asciinema v3 .cast file format reference. TRIGGERS - cast format, asciicast spec, event codes, parse cast file.

  • 75 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added February 7, 2026
toolsshellbashdebugginggitperformancedocumentation

Works with

  • claude code
  • terminal
  • cli

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add terrylica/cc-skills --skill asciinema-cast-format --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Asciinema Cast Format?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Asciinema Cast Format
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/terrylica-asciinema-cast-format/badge)](https://www.skillsdirectory.com/skills/terrylica-asciinema-cast-format)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: asciinema-cast-format
description: Asciinema v3 .cast file format reference. TRIGGERS - cast format, asciicast spec, event codes, parse cast file.
allowed-tools: Read, Bash
---

# asciinema-cast-format

Reference documentation for the asciinema v3 .cast file format (asciicast v2 specification).

> **Platform**: All platforms (documentation only)

## When to Use This Skill

Use this skill when:

- Parsing or inspecting .cast file structure
- Understanding NDJSON header and event formats
- Building tools that read or write .cast files
- Debugging recording issues or format errors
- Learning the asciicast v2 specification

---

## Format Overview

Asciinema v3 uses NDJSON (Newline Delimited JSON) format:

- Line 1: Header object with recording metadata
- Lines 2+: Event arrays with timestamp, type, and data

---

## Header Specification

The first line is a JSON object with these fields:

| Field       | Type   | Required | Description                                 |
| ----------- | ------ | -------- | ------------------------------------------- |
| `version`   | int    | Yes      | Format version (always 2 for v3 recordings) |
| `width`     | int    | Yes      | Terminal width in columns                   |
| `height`    | int    | Yes      | Terminal height in rows                     |
| `timestamp` | int    | No       | Unix timestamp of recording start           |
| `duration`  | float  | No       | Total duration in seconds                   |
| `title`     | string | No       | Recording title                             |
| `env`       | object | No       | Environment variables (SHELL, TERM)         |
| `theme`     | object | No       | Terminal color theme                        |

### Example Header

```json
{
  "version": 2,
  "width": 120,
  "height": 40,
  "timestamp": 1703462400,
  "duration": 3600.5,
  "title": "Claude Code Session",
  "env": { "SHELL": "/bin/zsh", "TERM": "xterm-256color" }
}
```

---

## Event Codes

Each event after the header is a 3-element array:

```json
[timestamp, event_type, data]
```

| Code | Name   | Description                 | Data Format          |
| ---- | ------ | --------------------------- | -------------------- |
| `o`  | Output | Terminal output (stdout)    | String               |
| `i`  | Input  | Terminal input (stdin)      | String               |
| `m`  | Marker | Named marker for navigation | String (marker name) |
| `r`  | Resize | Terminal resize event       | `"WIDTHxHEIGHT"`     |
| `x`  | Exit   | Extension for custom data   | Varies               |

### Event Examples

```json
[0.5, "o", "$ ls -la\r\n"]
[1.2, "o", "total 48\r\n"]
[1.3, "o", "drwxr-xr-x  12 user  staff  384 Dec 24 10:00 .\r\n"]
[5.0, "m", "file-listing-complete"]
[10.5, "r", "80x24"]
```

---

## Timestamp Behavior

- Timestamps are **relative to recording start** (first event is 0.0)
- Measured in seconds with millisecond precision
- Used for playback timing and navigation

### Calculating Absolute Time

```bash
/usr/bin/env bash << 'CALC_TIME_EOF'
HEADER_TIMESTAMP=$(head -1 recording.cast | jq -r '.timestamp')
EVENT_OFFSET=1234.5  # From event array

ABSOLUTE=$(echo "$HEADER_TIMESTAMP + $EVENT_OFFSET" | bc)
date -r "$ABSOLUTE"  # macOS
# date -d "@$ABSOLUTE"  # Linux
CALC_TIME_EOF
```

---

## Parsing Examples

### Extract Header with jq

```bash
/usr/bin/env bash << 'HEADER_EOF'
head -1 recording.cast | jq '.'
HEADER_EOF
```

### Get Recording Duration

```bash
/usr/bin/env bash << 'DURATION_EOF'
head -1 recording.cast | jq -r '.duration // "unknown"'
DURATION_EOF
```

### Count Events by Type

```bash
/usr/bin/env bash << 'COUNT_EOF'
tail -n +2 recording.cast | jq -r '.[1]' | sort | uniq -c
COUNT_EOF
```

### Extract All Output Events

```bash
/usr/bin/env bash << 'OUTPUT_EOF'
tail -n +2 recording.cast | jq -r 'select(.[1] == "o") | .[2]'
OUTPUT_EOF
```

### Find Markers

```bash
/usr/bin/env bash << 'MARKERS_EOF'
tail -n +2 recording.cast | jq -r 'select(.[1] == "m") | "\(.[0])s: \(.[2])"'
MARKERS_EOF
```

### Get Event at Specific Time

```bash
/usr/bin/env bash << 'TIME_EOF'
TARGET_TIME=60  # seconds
tail -n +2 recording.cast | jq -r "select(.[0] >= $TARGET_TIME and .[0] < $((TARGET_TIME + 1))) | .[2]"
TIME_EOF
```

---

## Large File Considerations

For recordings >100MB:

| File Size | Line Count | Approach                              |
| --------- | ---------- | ------------------------------------- |
| <100MB    | <1M        | jq streaming works fine               |
| 100-500MB | 1-5M       | Use `--stream` flag, consider ripgrep |
| 500MB+    | 5M+        | Convert to .txt first with asciinema  |

### Memory-Efficient Streaming

```bash
/usr/bin/env bash << 'STREAM_EOF'
# Stream process large files
jq --stream -n 'fromstream(1|truncate_stream(inputs))' recording.cast | head -1000
STREAM_EOF
```

### Use asciinema convert

For very large files, convert to plain text first:

```bash
asciinema convert -f txt recording.cast recording.txt
```

This strips ANSI codes and produces clean text (typically 950:1 compression).

---

## TodoWrite Task Template

```
1. [Reference] Identify .cast file to analyze
2. [Header] Extract and display header metadata
3. [Events] Count events by type (o, i, m, r)
4. [Analysis] Extract relevant event data based on user need
5. [Navigation] Find markers or specific timestamps if needed
```

---

## Post-Change Checklist

After modifying this skill:

1. [ ] Event code table matches asciinema v2 specification
2. [ ] Parsing examples use heredoc wrapper for bash compatibility
3. [ ] Large file guidance reflects actual performance characteristics
4. [ ] All jq commands tested with sample .cast files

---

## Reference Documentation

- [asciinema asciicast v2 Format](https://docs.asciinema.org/manual/asciicast/v2/)
- [asciinema CLI Usage](https://docs.asciinema.org/manual/cli/)
- [jq Manual](https://jqlang.github.io/jq/manual/)

---

## Troubleshooting

| Issue                   | Cause                        | Solution                                           |
| ----------------------- | ---------------------------- | -------------------------------------------------- |
| jq parse error          | Invalid NDJSON in .cast file | Check each line is valid JSON with `jq -c .`       |
| Header missing duration | Recording in progress        | Duration added when recording ends                 |
| Unknown event type      | Custom extension event       | Check for `x` type events (extension data)         |
| Timestamp out of order  | Corrupted file               | Events should be monotonically increasing          |
| Large file jq timeout   | File too big for in-memory   | Use `--stream` flag or convert to .txt first       |
| Markers not found       | No markers in recording      | Markers are optional; not all recordings have them |
| Wrong version number    | Older cast format            | This skill covers v2 format (asciinema v3+)        |
| Empty output from tail  | File has only header         | Recording may be empty or single-line              |

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.