session-export

Produces a one-page, fully offline HTML view of a Claude Code or OpenAI Codex session — user/assistant turns, thinking/reasoning, tool calls and results, and any subagent transcripts rendered inline.

When to invoke

• User asks to "export the current session", "save this chat", "make an HTML of this conversation".
• User wants to share a session with someone who doesn't have the CLI installed.

How to run

Two exporters sit next to this SKILL.md — use the one matching the CLI you're running in. Use that directory's path (it may be project-local or under ~/.claude/ or ~/.codex/, so don't hardcode a project-relative path):

# Claude Code
node <this-skill-dir>/export.js [session-id-or-path] [--no-redact]
# OpenAI Codex
node <this-skill-dir>/codex-export.js [session-id-or-path] [--no-redact]

Run from the working directory of the session you want to export. Auto-detection:

• Claude Code (export.js): newest main session jsonl under <configDir>/projects/<encoded-cwd>/, where configDir is $CLAUDE_CONFIG_DIR if set, else ~/.claude (skips agent-*.jsonl siblings and empty stubs).
• OpenAI Codex (codex-export.js): newest rollout-*.jsonl under <codexHome>/sessions/, where codexHome is $CODEX_HOME if set, else ~/.codex.

Both accept a session id or an explicit path as the argument, and --no-redact to disable the secret-redaction pass (see below).

The script writes claude-session-… / codex-session-…_<sessionId>.html to the current working directory and prints the absolute path. When secrets are redacted, a one-line summary is also written to stderr.

Secret redaction

Before the HTML is written, every message text/thinking block, every tool input (recursively), and every tool result is scanned for well-known credential shapes and matches are replaced with [REDACTED:<kind>]:

• sk-ant-… (Anthropic), sk-… / sk-proj-… (OpenAI), sk_live_… / pk_live_… / sk_test_… (Stripe)
• AKIA… / ASIA… / AIDA… (AWS access keys)
• ghp_… / ghs_… / gho_… / ghu_… / ghr_… (GitHub PATs)
• xoxb-… / xoxp-… / xoxa-… / xoxs-… (Slack)
• AIza… (Google API keys), eyJ…eyJ….… (JWTs)
• -----BEGIN … PRIVATE KEY----- blocks

Counts per kind are stored on header.redactions and surfaced as a yellow redacted:<n> chip in the exported header (hover for the breakdown). This is a heuristic pass — review your exports before sharing.

Output

Single HTML file with:

• Embedded JSON payload (<script id="session-data" type="application/json">), no base64.
• Inline CSS supporting light + dark themes (16px base, em-based), respects prefers-color-scheme, manual toggle persisted in localStorage.
• Sidebar tree of turns with live search and filter chips.
• Collapsible thinking blocks, collapsible tool outputs.
• Subagent (Task/Agent) cards that expand to show the nested subagent transcript (loaded from sibling agent-<id>.jsonl files when present, falls back to inline progress records).
• Keyboard shortcuts: / focus search, t toggle thinking, o toggle tool outputs, b toggle sidebar, a/z expand/collapse all, Esc clear search, [/] navigate entries.

No network access required to view.