Asciinema Cast Format

asciinema-cast-format

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

Platform: All platforms (documentation only)

Self-Evolving Skill: This skill improves through use. If instructions are wrong, parameters drifted, or a workaround was needed — fix this file immediately, don't defer. Only update for real, reproducible issues.

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:

Example Header

{
  "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:

[timestamp, event_type, data]

Event Examples

[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

/​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

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

Get Recording Duration

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

Count Events by Type

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

Extract All Output Events

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

Find Markers

/​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

/​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:

Memory-Efficient Streaming

/​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:

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
• asciinema CLI Usage
• jq Manual

---

Troubleshooting

Post-Execution Reflection

After this skill completes, check before closing:

1. Did the command succeed? — If not, fix the instruction or error table that caused the failure.
2. Did parameters or output change? — If the underlying tool's interface drifted, update Usage examples and Parameters table to match.
3. Was a workaround needed? — If you had to improvise (different flags, extra steps), update this SKILL.md so the next invocation doesn't need the same workaround.

Only update if the issue is real and reproducible — not speculative.