A skill that converts a coding-agent session transcript into a single self-contained HTML file — user/assistant turns, thinking/reasoning, every tool call with its result, and any subagent transcripts rendered inline.
Works with two CLIs, each reading the local .jsonl transcripts that CLI
writes:
- Claude Code —
export.js, reading~/.claude/projects/<encoded-cwd>/. - OpenAI Codex —
codex-export.js, reading~/.codex/sessions/….
Both run offline against on-disk session files; the desktop/web/IDE clients of either tool don't produce these files.
It ships both as a Claude Code plugin/skill (.claude-plugin/) and an
OpenAI Codex plugin/skill (.codex-plugin/), with the skill itself in
skills/session-export/.
Claude Code — copy the skill into a project, or symlink it user-globally:
cp -R skills/session-export /path/to/other-project/.claude/skills/
# or
ln -s "$PWD/skills/session-export" ~/.claude/skills/session-exportOpenAI Codex — the repo is a Codex plugin (.codex-plugin/plugin.json).
Add it to a Codex marketplace and install, or point Codex at this repo as a
local plugin source. The bundled session-export skill is then available in
Codex sessions.
In Claude Code, run /session-export, or call the matching exporter directly:
# Claude Code session
node skills/session-export/export.js [session-id-or-path] [--no-redact]
# OpenAI Codex session
node skills/session-export/codex-export.js [session-id-or-path] [--no-redact]- no argument — most recently modified session for the current context
- session id — looked up in the CLI's session store
- path — used directly
--no-redact— skip the credential-redaction pass
The HTML is written to claude-session-… / codex-session-…_<sessionId>.html
in the current directory and its absolute path printed to stdout.
By default every message, thinking block, tool input, and tool result
is scanned for well-known credential shapes (Anthropic / OpenAI / Stripe
keys, AWS access keys, GitHub PATs, Slack tokens, Google API keys, JWTs,
-----BEGIN … PRIVATE KEY----- blocks) and matches are replaced with
[REDACTED:<kind>]. Counts appear as a yellow redacted:<n> chip in the
header. It's a heuristic pass — review exports before sharing.
- Embedded JSON payload in
<script id="session-data">(no base64). - Inline CSS, no CDN. Searchable sidebar tree with filter chips (user / asst / tools / subagent); resizable, collapsible.
- Collapsible thinking blocks and tool outputs; subagent
(
Task/Agent/Explore/ …) cards expand to the full nested transcript, recursively. - Keyboard:
/focus search ·tthinking ·otool outputs ·bsidebar ·[/]prev/next turn ·Escclear search.
Ships dark (default) and light, respects prefers-color-scheme, and
the toggle is persisted in localStorage. Add more by dropping a
themes.css next to template.js: each :root[data-theme="<name>"]
block it defines is appended to the export and added to the toggle
cycle — omitted variables inherit from dark. The shipped themes.css
includes a cool example.
Claude Code has shipped three transcript layouts; the exporter handles
all three: per-session <sessionId>/subagents/agent-<id>.jsonl (+
.meta.json); agent-<id>.jsonl siblings linked by sessionId; and
inline progress records linked by parentToolUseID (deduped by inner
uuid). External tool-result spillover under
<sessionId>/tool-results/<id>.txt is inlined when present.
.claude-plugin/plugin.json — Claude Code plugin manifest (+ marketplace.json)
.codex-plugin/plugin.json — OpenAI Codex plugin manifest
.agents/plugins/marketplace.json — Codex marketplace entry
skills/session-export/
SKILL.md — skill manifest (used by both CLIs)
export.js — Claude Code session resolver + payload builder
codex-export.js — OpenAI Codex rollout resolver + payload builder
redact.js — shared secret-redaction pass
template.js — inline CSS + scaffold + client renderer (shared)
themes.css — optional extra themes (cool example)
No runtime dependencies; Node 18+.
Open source under the MIT license. No tracking. No analytics.