| name | cortex-replay |
| description | Generate interactive HTML replays from Cortex Code sessions. Use when: user wants to create a replay, share a session, export a session as HTML, record a demo, make a screencast, list sessions. Triggers: replay, session replay, share session, export session, HTML replay, cortex-replay, record session, session demo, shareable session. |
| tools | ["Bash","Read","Write"] |
cortex-replay
Generate self-contained, interactive HTML replays from Cortex Code session transcripts using the bundled Python script.
Prerequisites
None. This skill is fully self-contained -- it uses only Python 3 stdlib (no external dependencies, no npm, no Node.js).
The replay script is bundled at <SKILL_DIR>/scripts/replay.py. Run it directly using its absolute path -- no installation needed.
Determine the script path by resolving <SKILL_DIR> to the actual skill directory. For local skills this is typically ~/.cortex/skills/cortex-replay/scripts/replay.py or ~/.snowflake/cortex/skills/cortex-replay/scripts/replay.py. For stage-deployed skills, check the cached path.
Quick check:
python3 <SKILL_DIR>/scripts/replay.py --help
Workflow
1. Determine what the user wants
Ask if needed:
- Which session? Options:
--last (most recent), a session ID, or a file path
- Output file? Default:
replay.html in the current working directory
- Any customizations? Theme, turn range, bookmarks, speed, hide thinking/tools
If the user just says "create a replay" or "replay my session" without specifics, default to --last with the snowflake theme.
2. List sessions (if user needs to pick one)
python3 <SKILL_DIR>/scripts/replay.py --list-sessions
This shows all sessions with IDs, titles, timestamps, and turn counts. Help the user pick the right one. Partial ID matching works (e.g., 6868f059 matches 6868f059-3b8a-423d-8a51-ef0397c7f469).
3. Generate the replay
Build the command from user preferences. Base command:
python3 <SKILL_DIR>/scripts/replay.py <input> -o <output.html>
Input options (mutually exclusive):
--last -- most recent session<session-id> -- partial or full session ID<path/to/session.json> -- direct file path
Customization flags:
| Flag | What it does | Example |
|---|
--theme NAME | Visual theme | --theme tokyo-night |
--turns N-M | Include only turns N through M | --turns 3-15 |
--from TIMESTAMP | Start time filter (ISO 8601) | --from "2026-03-01T10:00" |
--to TIMESTAMP | End time filter (ISO 8601) | --to "2026-03-01T12:00" |
--speed N | Initial playback speed (0.5-5) | --speed 2 |
--no-thinking | Hide thinking blocks by default | |
--no-tool-calls | Hide tool call blocks by default | |
--no-redact | Disable automatic secret redaction | |
--no-animate | Disable typewriter animation (classic mode) | |
--title TEXT | Custom page title | --title "Bug Fix Demo" |
--mark "N:Label" | Bookmark at turn N (repeatable) | --mark "3:Setup" --mark "8:Fix" |
--bookmarks FILE | JSON file with bookmarks | --bookmarks marks.json |
--theme-file FILE | Custom theme JSON | --theme-file my-theme.json |
--user-label NAME | Label for user messages | --user-label "Developer" |
--assistant-label NAME | Label for assistant messages | --assistant-label "Coco" |
--no-compress | Embed raw JSON (larger file, debuggable) | |
Available themes: snowflake (default), tokyo-night, monokai, solarized-dark, github-light, dracula
4. Open the replay
After generating, open it:
open <output.html>
5. Report results
Tell the user:
- Output file path and size
- Number of turns included
- Theme used
- How to share it (it's fully self-contained -- email, embed, commit to repo)
Example Commands
python3 <SKILL_DIR>/scripts/replay.py --last -o replay.html
python3 <SKILL_DIR>/scripts/replay.py 6868f059 --theme dracula -o replay.html
python3 <SKILL_DIR>/scripts/replay.py --last --turns 3-15 --theme tokyo-night --speed 2 \
--no-thinking --mark "3:Setup" --mark "8:Implementation" \
--title "Building a Data Pipeline" -o demo.html
python3 <SKILL_DIR>/scripts/replay.py --last --theme github-light --no-tool-calls -o docs-replay.html
python3 <SKILL_DIR>/scripts/replay.py session.json --user-label "Engineer" --assistant-label "Coco" -o replay.html
Embedding in Docs or Blogs
The output HTML can be embedded via iframe:
<iframe src="replay.html" width="100%" height="600"
style="border: 1px solid #333; border-radius: 8px;"></iframe>
Player Controls (for the user's reference)
The generated HTML player supports:
- Typewriter animation -- user prompts are typed out character-by-character with a blinking cursor; assistant responses fade in block-by-block (disable with
--no-animate)
- Play/Pause -- auto-advances through turns with animated transitions
- Step forward/back -- jump between turns (instantly reveals content when navigating manually)
- Progress bar -- click to jump anywhere (instantly reveals all content up to that point)
- Speed control -- 0.5x to 5x (scales typing speed and block reveal timing)
- Toggle checkboxes -- show/hide thinking and tool calls
- Keyboard: Space/K = play/pause, Right/L = forward, Left/H = back
Custom Themes
Users can create a JSON file with any subset of these keys (missing keys fall back to snowflake defaults):
{
"bg": "#1a1a2e",
"bg-surface": "#16213e",
"bg-hover": "#1a2744",
"text": "#e0e0e0",
"text-dim": "#666",
"text-bright": "#fff",
"accent": "#e94560",
"accent-dim": "#b33548",
"green": "#4ecca3",
"blue": "#4cc9f0",
"orange": "#f0a030",
"red": "#e85454",
"cyan": "#4cc9f0",
"border": "#2a2a4a",
"tool-bg": "#12121f",
"thinking-bg": "#111122"
}
Use with: --theme-file my-theme.json
Session File Location
Cortex Code sessions are stored at ~/.snowflake/cortex/conversations/. Each .json file is a full session transcript.
Notes
- The output HTML is fully self-contained -- no external dependencies, works offline
- Secrets (API keys, tokens, passwords, Snowflake credentials) are automatically redacted unless
--no-redact is used
- System-reminder blocks and internal platform noise are stripped automatically
- The
--last flag reads ~/.snowflake/cortex/conversations/.last-session to find the most recent session
- This skill requires only Python 3 (stdlib) -- no npm, no Node.js, no pip packages
Deploying to a Profile
To publish this skill to a Snowflake stage for profile-based distribution:
cortex skill publish ~/.cortex/skills/cortex-replay --to-stage @DB.SCHEMA.STAGE/skills/
This uploads both SKILL.md and scripts/replay.py to the stage. Any user with the profile will get the skill with the bundled script -- no installation required.
