← All reportsPublic report

garrytan/gstack

Use Garry Tan's exact Claude Code setup: 23 opinionated tools that serve as CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA

130k stars TypeScript View on GitHubprofiled 11d ago
Try GitZoid
17PRs this week
1Contributors
12Deps scanned
0Issues found
01 · Repo overview

How gstack is put together

gstack is a Claude Code 'skills' distribution: a large collection of Markdown-defined slash-command specialists plus a TypeScript/Bun headless-browser toolkit ('browse') that backs the QA/review skills. The repo is primarily TypeScript run on Bun, with a Chrome extension (vanilla JS) that pairs with a local browse HTTP server, and ~80 small Bash/TS helper binaries under bin/. Skills are authored once as SKILL.md.tmpl and generated per AI-agent host (Claude, Codex, Cursor, etc.) via a host-config registry in hosts/. The browse server keeps per-project state in a gitignored .gstack/ directory, exposes an auth-token-gated command API to the extension, and includes security layers for untrusted web content (datamarking, hidden-element stripping, trust envelopes) and an opt-in ngrok pair-agent remote-sharing mode.

Languages

TypeScriptJavaScriptMarkdownBash

Frameworks

PlaywrightChrome Extension (Manifest V3 service worker)@anthropic-ai/claude-agent-sdkReact (diagram-render bundle)Mermaid/Excalidraw (offline diagram HTML)

Infrastructure

ngrok tunnel (@ngrok/ngrok) for opt-in pair-agent remote browser sharingGitLab CI (.gitlab-ci.yml)GitHub-hosted distribution via git clone into ~/.claude/skillsClawHub skill publishing

Major components

browse CLI + server daemon

Headless Playwright browser automation exposed as a local HTTP server and $B CLI commands, with per-project state in .gstack/.

Cookie picker / cookie import

HTTP routes and browser-profile logic that decrypt and import cookies from installed browsers into the Playwright session, with one-time-code auth and v20 App-Bound-Encryption CDP fallback.

Content security layer

Defense-in-depth against web-content prompt injection: datamarking watermarks, hidden-element/ARIA-injection detection, untrusted-content envelopes, and pluggable content filters.

Domain skills storage

Per-host learned markdown skills stored in project/global JSONL with quarantine states, versioning, rollback, and classifier-gated promotion.

Chrome extension

Sidepanel/popup/content-script companion that talks to the browse server, with sender-auth gating of privileged token-bearing messages.

Host registry (hosts/)

Per-agent host configs (claude, codex, cursor, openclaw, etc.) used by setup to install generated skills to each agent's skills directory.

Skills library (*/SKILL.md.tmpl)

~40 Markdown slash-command workflows (/review, /qa, /cso, /ship, /autoplan, ...) templated per host by gen-skill-docs.

bin/ operational tooling

Dozens of small Bash/TS utilities for config (~/.gstack/config.yaml), telemetry, egress receipts, redaction, brain sync, detach/locking, and team bootstrap.

Over the past month the gstack project ran a sustained cleanup-and-hardening campaign: large 'tracker waves' closed hundreds of stale issues and pull requests while absorbing dozens of community contributions with documented proof. Alongside that, the team fixed a series of silent-failure bugs where safety checks weren't actually working, tightened security for outside contributors, and shipped performance and reliability improvements like faster test suites and content-integrity checks.

Week by week

2026-08-24A quieter week focused on closing out the silent-failure cleanup wave with six fixes and five community contributions.latest1 change

Fix

Silent-failure wave completed

Six fixes landed so that checks which used to fail quietly now fail loudly and visibly, with five community-submitted improvements folded in and the tracking issue closed with evidence.

2026-08-17Seven releases delivered security hardening, tracker cleanup of over 100 stale items, and fixes to pairing and tunnel revocation.6 changes

Fix

Re-pairing now revokes old access immediately

When you re-pair a device to narrow its permissions, the previous broader access grant is revoked on the spot instead of lingering.

Fix

Tunnel revocation verified end-to-end

Revoking a tunnel now genuinely cuts off everything tied to it, including setup keys, confirmed by live testing.

Fix

Phantom hooks cleaned up

Stray AskUserQuestion hooks no longer register from unexpected places, and settings files can repair themselves when they drift.

Fix

Second tracker wave closes 111 stale items

Sixteen verified fixes went in while 90 outdated pull requests and 21 issues were closed, each with documented justification.

Feature

Bounded-scope profile for Codex installs

A new gpt-5.6-sol profile lets Codex setups run with tighter, safer limits out of the box.

Fix

External-contributor security sweep

Six security findings in community-contributed code were hardened and locked in with regression tests so they can't come back.

2026-08-10The busiest week yet: seven releases including a massive audit that closed ~50 community PRs, a code-smell purge removing nearly 25,000 lines, and major speedups to testing infrastructure.6 changes

Feature

Content binding and trust envelope

New integrity features track an evidence ledger, detect stale worktrees, and make hooks fail closed (secure by default) rather than open.

Refactor

Faster, more honest test suite

The free test suite now runs truthfully in about 90 seconds, evaluations are billed only on changed code, and Linux became a required test lane.

Feature

Fork port wave 2

Feature fixes, session persistence, Apple release support, and supply-chain checks were ported over from the fork.

Refactor

Code-smell purge removes ~25k lines

Every pipeline guard was reworked so it provably fires when it should, shrinking the codebase by nearly 25,000 lines.

Fix

Full tracker audit closes 52 issues

Ninety fixes landed and 52 issues were closed, absorbing roughly 50 community pull requests along the way.

Feature

Plan-mode auto-select at review gate

The review scope gate now automatically picks plan mode where appropriate, reducing manual setup.

2026-08-03Two releases kicked off the reliability push by fixing guards that failed silently or openly when they shouldn't.2 changes

Fix

Guards no longer fail open

Nine fixes ensured safety checks stop passing through silently when something goes wrong, with four community PRs absorbed.

Fix

Free-suite drift on dev machines fixed

Corrected path and pattern-matching issues so local development machines get consistent evaluation results, plus a baseline observability floor.

03 · Security check

Dependencies and code review

12 dependencies scanned

Dependency advisories

Security Watch

No known advisories across 12 scanned dependencies.

No known advisories in the scanned dependencies.

Code review

No concrete code-level issues confirmed.

Get this report every week for your repos.

GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.

$19 a month, flat · First 10 outputs free · No card required