/sx - Session Export (Multi-LLM Interactive)

Export LLM session data to CODITECT Universal Session Format (CUSF) with interactive multi-select capability. Supports Claude, Codex, Gemini, and Kimi. Pairs with /cx for the complete export → extract workflow.

Usage

# Interactive mode (default) - checkbox multi-select
/sx

# Export specific LLMs (non-interactive)
/sx --llm claude
/sx --llm codex,gemini
/sx --llm all

# Show status only (no export)
/sx --status

# Dry run (show what would be exported)
/sx --dry-run

# Export and immediately process with /cx
/sx --process
/sx -p

# List available sessions
/sx --list
/sx --list --llm codex

# Export past sessions
/sx --session <session-id>
/sx --source ~/.claude/projects/abc/def.jsonl

# Output options
/sx --output /path/to/export.jsonl
/sx --format jsonl|json|sqlite

# Reconstruction mode (includes ALL metadata)
/sx --reconstruct

System Prompt

⚠️ EXECUTION DIRECTIVE: When the user invokes /sx with ANY arguments (or no arguments), you MUST:

IMMEDIATELY execute the command - no questions, no explanations first
ALWAYS show full output from the script execution
ALWAYS provide summary after execution completes

DO NOT:

Say "I don't need to take action" - you ALWAYS execute when invoked
Ask for confirmation - the user invoking the command IS the confirmation
Skip execution even if it seems redundant - run it anyway

Workflow: /sx (export) → /cx (extract & index)

ADR-159 Project Scoping: When --project is provided (or auto-detected from $CODITECT_PROJECT or CWD), pass it to the extractor via --project flag so exported sessions are attributed to the correct project. This ensures /cx processes them into the right project scope.

# With project scope:
python3 ~/.coditect/scripts/unified-message-extractor.py --project PILOT [files...]

Output Directories (ADR-122 naming convention):

Claude: ~/PROJECTS/.coditect-data/sessions-export-pending-anthropic/
Codex: ~/PROJECTS/.coditect-data/sessions-export-pending-codex/
Gemini: ~/PROJECTS/.coditect-data/sessions-export-pending-gemini/
Kimi: ~/PROJECTS/.coditect-data/sessions-export-pending-kimi/

Interactive Mode (Default - No Arguments)

When invoked without --llm, use the AskUserQuestion tool to present an interactive multi-select:

{
  "questions": [{
    "question": "Which LLM sessions do you want to export?",
    "header": "Export LLMs",
    "multiSelect": true,
    "options": [
      {"label": "Claude (anthropic)", "description": "Export current Claude session - XX% context"},
      {"label": "Codex", "description": "Export current Codex session - XX% context"},
      {"label": "Gemini", "description": "Export current Gemini session - XX% context"},
      {"label": "Kimi", "description": "Export current Kimi session - XX% context"}
    ]
  }]
}

Before presenting options:

Check which LLMs are running: pgrep -l "claude|codex|gemini|kimi"
Get context usage for each running LLM
Include status in option descriptions (e.g., "89% context ⚠️" or "Not running")

After selection: Execute the export for each selected LLM using the existing extractors:

# For each selected LLM, run its extractor:
PENDING_BASE=~/PROJECTS/.coditect-data

# Claude
python3 ~/.coditect/scripts/extractors/claude_extractor.py ~/.claude/projects/ \
  --output "$PENDING_BASE/sessions-export-pending-anthropic/"

# Codex
python3 ~/.coditect/scripts/extractors/codex_extractor.py ~/.codex/sessions/ \
  --output "$PENDING_BASE/sessions-export-pending-codex/"

# Gemini
python3 ~/.coditect/scripts/extractors/gemini_extractor.py ~/.gemini/tmp/ \
  --output "$PENDING_BASE/sessions-export-pending-gemini/"

# Kimi
python3 ~/.coditect/scripts/extractors/kimi_extractor.py ~/.kimi/sessions/ \
  --output "$PENDING_BASE/sessions-export-pending-kimi/"

Non-Interactive Mode (--llm specified)

Execute directly without prompting using existing extractors:

# Single LLM
python3 ~/.coditect/scripts/extractors/claude_extractor.py ~/.claude/projects/ \
  --output ~/PROJECTS/.coditect-data/sessions-export-pending-anthropic/

# Multiple LLMs (comma-separated)
for llm in claude codex; do
  python3 ~/.coditect/scripts/extractors/${llm}_extractor.py ...
done

Options

Option	Short	Description
`--llm LLM`		LLM(s) to export: `claude`, `codex`, `gemini`, `kimi`, `all`, or comma-separated
`--status`	`-s`	Show LLM status only (running, context %, session info)
`--process`	`-p`	Run `/cx` after export to process pending files
`--list`	`-l`	List available sessions
`--source PATH`		Specific session file to export
`--session-id ID`		Specific session ID to extract
`--output PATH`	`-o`	Output file path (default: pending directory)
`--format FMT`	`-f`	Output format: jsonl, json, sqlite
`--pending`		Write to pending directory (default for interactive)
`--compress`		Gzip compress output
`--reconstruct`		Include all metadata for reconstruction
`--dry-run`		Preview without writing
`--force`		Export even if context usage is low
`--detect`		Show detected LLM CLI tools
`--project ID`		Project scope for export attribution (ADR-159). Auto-detected from CWD if not set.
`--verbose`	`-v`	Verbose output

Interactive Display

╔══════════════════════════════════════════════════════════════╗
║                    SESSION EXPORT                            ║
╠══════════════════════════════════════════════════════════════╣
║ LLM Status:                                                  ║
║ ├─ Claude:  Running  42% context  ✓                          ║
║ ├─ Codex:   Running  89% context  ⚠️ (high)                   ║
║ ├─ Gemini:  Running  31% context  ✓                          ║
║ └─ Kimi:    Not running                                      ║
╠══════════════════════════════════════════════════════════════╣
║ Selected: Claude, Codex                                      ║
║ Exporting...                                                 ║
║ ├─ Claude → sessions-export-pending-anthropic/               ║
║ │   ✓ 156 messages (2.3 MB)                                  ║
║ └─ Codex → sessions-export-pending-codex/                    ║
║     ✓ 423 messages (5.1 MB)                                  ║
╠══════════════════════════════════════════════════════════════╣
║ Total: 579 messages exported to 2 pending directories        ║
║                                                              ║
║ Next: Run /cx to process pending exports                     ║
╚══════════════════════════════════════════════════════════════╝

CUSF Output Format

Line-delimited JSON in CODITECT Universal Session Format:

{"_meta": {"format": "cusf", "version": "1.0.0", "exported_at": "2026-01-29T08:00:00Z"}}
{"type": "session_start", "session_id": "uuid", "llm_source": "claude", "llm_model": "claude-opus-4-5", ...}
{"type": "message", "role": "user", "content": "...", "timestamp": "...", ...}
{"type": "message", "role": "assistant", "content": "...", "timestamp": "...", ...}
{"type": "tool_use", "tool_name": "Read", "tool_input": {...}, ...}
{"type": "tool_result", "tool_id": "...", "result": "...", ...}
{"type": "session_end", "session_id": "uuid", "total_messages": 42, ...}

Workflow Integration

┌─────────────┐     ┌─────────────────────────────┐     ┌─────────────┐
│   /sx       │────▶│    Pending Directories      │────▶│   /cx       │
│  (export)   │     │                             │     │  (extract)  │
└─────────────┘     │ sessions-export-pending-    │     └─────────────┘
                    │   anthropic/                │
                    │   codex/                    │
                    │   gemini/                   │
                    │   kimi/                     │
                    └─────────────────────────────┘

One-step workflow:

/sx -p                    # Export + process in one command

Two-step workflow:

/sx && /cx                # Export, then extract

Examples

Interactive Multi-Select (Default)

/sx

Result: Shows multi-select dialog, exports selected LLMs to pending directories

Export All LLMs

/sx --llm all

Result: Exports all 4 LLMs without interaction

Export and Process Immediately

/sx --llm claude -p

Result: Exports Claude session, then runs /cx to extract messages

Check Status Only

/sx --status

Result: Shows which LLMs are running and their context usage

List Codex Sessions

/sx --list --llm codex

Result: Lists available sessions in Codex history

Export to SQLite

/sx --llm claude --format sqlite --output ~/backup.db

Result: Exports to portable SQLite database

Detect Installed Tools

/sx --detect

Result: Shows which LLM CLI tools are installed

Differences from /export

Feature	`/export` (built-in)	`/sx`
LLM Support	Claude only	Claude, Codex, Gemini, Kimi
Interactive	No	Yes (multi-select)
Format	TXT	JSONL, JSON, SQLite (CUSF)
Reconstruction	No	Yes
Past Sessions	No	Yes (`--session-id`)
Cross-LLM	No	Yes (unified format)
Pending Integration	No	Yes (writes to pending dirs)

Command	Purpose
`/cx`	Process pending exports, extract messages to database
`/cxq`	Query extracted messages
`/export`	Built-in Claude export (text format)

Success Output

✅ COMMAND COMPLETE: /sx
LLMs Exported: Claude, Codex
Messages: 579 total
Pending: 2 directories ready for /cx
Next: Run /cx to process

Completion Checklist

Failure Indicators

This command has FAILED if:

❌ No LLMs running (and none specified)
❌ Write permission denied to pending directory
❌ Extractor script failed
❌ No sessions to export

When NOT to Use

Do NOT use when:

Just want to process existing pending files (use /cx instead)
Want text export only (use /export built-in)
LLM is in middle of critical operation

Anti-Patterns (Avoid)

Anti-Pattern	Problem	Solution
Export without /cx	Pending files accumulate	Use `/sx -p` or run `/cx` after
Export low-context sessions	Unnecessary work	Check `--status` first
Ignore high-context warnings	Risk context overflow	Export soon when ⚠️ shown

Principles

This command embodies:

#3 Complete Execution - Full export pipeline
#6 Clear, Understandable - Interactive selection with status
#1 Recycle, Extend - Leverages existing extractors

Full Standard: CODITECT-STANDARD-AUTOMATION.md

Script: .coditect/scripts/session-export.py Version: 2.0.0 Created: 2026-01-28 Updated: 2026-01-29 Track: J.13 (Memory - Generic Session Export) Related ADRs: ADR-122 (Unified LLM Architecture), ADR-128 (CUSF Format)

Usage​

System Prompt​

Interactive Mode (Default - No Arguments)​

Non-Interactive Mode (--llm specified)​

Options​

Interactive Display​

CUSF Output Format​

Workflow Integration​

Examples​

Interactive Multi-Select (Default)​

Export All LLMs​

Export and Process Immediately​

Check Status Only​

List Codex Sessions​

Export to SQLite​

Detect Installed Tools​

Differences from /export​

Related Commands​

Success Output​

Completion Checklist​

Failure Indicators​

When NOT to Use​

Anti-Patterns (Avoid)​

Principles​