LinkedIn    Email    agent-tree

I'm an AI engineer and co-founder at SECA, an AI startup spun out of LG Electronics, based in Seoul.

I design end-to-end AI pipelines and turn them into products people can use: data processing, models, tools and evaluation, connected into one system. I use AI throughout development to build, test and refine that work. At SECA, I apply this approach to complex hardware data.

Before SECA I worked at LG Electronics AI LAB and NAVER, and did research at Seoul National University.

Navigate Claude Code and Codex sessions as numbered trees, and pick up from any point.

A long agent session is hard to look back on. agent-tree reads the session logs that Claude Code and Codex CLI already keep and turns each one into a numbered outline: every prompt you typed, with the reads, edits and commands it triggered nested underneath. Pick any step and you get a ready-to-paste prompt that continues from there in a fresh session, or forks to try another way.

• In the terminal: an interactive tree, plus --list, --diff and --json output for scripts.
• Inside Claude Code: a plugin with six MCP tools, so Claude can list, browse and resume sessions for you.
• Local and redacted: it reads the logs already on your machine and strips API keys, tokens and card numbers before anything is shown, copied or exported.

npm install -g @seungwoolee/agent-tree

Repository  ·  npm package  ·  Latest release

If it saves you a long scroll through a session log, a star on the repo helps other people find it.

I fix the reliability bugs I hit in the AI developer tools I use every day: orphaned processes, crashes, silent failures and lost data. 36 of my changes have been merged into 14 projects with more than 200,000 stars combined.

Stars	Project	PRs	Contribution
	oh-my-openagent code-yeongyu	1 #7525	Stopped abandoned stdio MCP servers from running forever: the idle timeout never closed stdin, so they could not exit. Reported the bug and wrote the fix.
	oh-my-claudecode Yeachan-Heo	9 #1843 #1480 #1482	Unified 8 divergent process-liveness checks, 5 of which mistook live processes for dead ones on EPERM, and wired stop-hook safeguards into the scripts that actually run instead of code that was never invoked.
	oh-my-codex Yeachan-Heo	6 #1605 #1918 #1303	Prevented orphaned codex processes when tmux kills a session, kept the HUD and dispatch loop alive through partial writes, and made failed state writes fail loudly instead of reporting success. Also reported three high-severity dependency vulnerabilities, since patched.
	cmux manaflow-ai	1 #1834	Removed force-unwraps from file-write paths, switching to UTF-8 conversions that cannot fail.
	openworker andrewyng	1 #70	Made conversation-log rewrites atomic, closing a window where a crash during context compaction could wipe the history.
	opencodex lidge-jun	2 #3097 #3137	Capped the response spill directory, which was unbounded and could fill a disk within an hour, and fixed tests that failed whenever pid 4242 happened to be in use.
	PyMuPDF pymupdf	1 #5139	Turned a hard segfault in the C++ bindings into a reported error: an exception in a get_cdrawings() callback used to crash the process, and extraction now carries on.
	ouroboros Q00	3 #1152 #325 #154	Stopped the welcome suggestion from reappearing in every session for already-configured users, fixed a crash on null MCP parameters, and narrowed 30 bare except Exception clauses that could hide real bugs.
	gajae-code Yeachan-Heo	6 #5422 #6260 #6333	Made quota failover mark the account that actually ran out, recovered OAuth tokens rejected before expiry that left quota readings stale, and stopped queued requests from receiving another run's answer. Also reported a rate limit hidden behind a generic stream timeout, since fixed.
	oh-my-agent first-fluke	1 #265	Fixed an SQL-injection check that had never actually run: broken shell quoting made it pass every input, including real injection patterns.
	prism-insight dragon1086	1 #227	Fixed a module-cache collision that crashed the US stock analysis pipeline with import errors.
	openlogs charlietlamb	1 #3	Added a pre-publish gate, so a failing test suite can no longer ship a broken package to npm.
	memkraft seojoonkim	2 #3 #2	Fixed recall reporting every stored entity as missing, so agents see their stored memories again, and made track() return the path it creates.
	agf subinium	1 #1	Removed duplicated path helpers from the delete command, so it resolves the Claude and Codex directories through the shared config code.

• AI product engineering: pipeline design, model integration, evaluation and workflow automation.
• AI for hardware data: applying end-to-end AI pipelines to complex hardware data at SECA.
• Developer tools for AI-assisted work: practical tools, like agent-tree, that make agent sessions easier to review and resume.
• Quantitative research: applying AI to data-driven research and systematic experimentation.

You can reach me on LinkedIn or by email. For agent-tree bugs and ideas, open an issue.