How to Export Clean Session Transcripts
Why agent audits fail on collapsed tool output, and the exact commands, settings, and file paths to get 100% fair verification in Second Opinion.
When you copy from a terminal or chat UI that collapses tool output, the auditor only sees summary labels:
Agent: I'll run the test suite and verify the fix. [Ran 3 commands] Agent: All tests pass! Ready to ship.
When you use native session export, the actual command invocations and outputs are fully preserved:
Agent: Running verification tests. $ npm test PASS tests/auth.test.ts (14 tests passed, 0 failures) Exit code: 0 Agent: All tests pass! Ready to ship.
Choose your agent harness
Claude Code
Anthropic's official terminal agent CLI
How to export the full transcript
1. Use the native /export command (Recommended)
Inside any active Claude Code terminal session, type /export followed by a file path. Claude Code dumps the complete transcript with full, expanded tool calls directly to disk.
2. Copy the dumped markdown file
Open session-audit.md in your editor or print it to your terminal with cat/Get-Content, then copy it into Second Opinion.
3. Direct JSONL project logs
Claude Code stores raw structured execution logs locally under ~/.claude/projects/. Each file contains the full history of model calls, bash executions, and file diffs.
Copying directly from the terminal pane while tools are running. Claude Code collapses completed commands into 'Ran 2 commands' or 'used 3 tools' on screen. Second Opinion flags these as unverified because the actual command stdout is missing.
Always run `/export session.md` before ending your task. It guarantees 100% evidentiary completeness for Second Opinion audits.
Ready with your transcript? Run an audit in Second Opinion to verify claims and get paste-ready repair prompts.
Audit in Second Opinion →Paste your transcript to test for collapsed markers
We test your clipboard client-side for collapsed headers like “Ran 2 commands” before you submit it for an audit.