Garry Tan

Watches a fresh deploy for anomalies in latency, error rate, and user behavior. Flags regressions before they scale, with metrics and logs pinpointed to the root.

Spins up a headless browser to click through user flows, check for broken links, and verify forms work end-to-end without leaving Claude.

Chains together CEO, design, engineering, and DX reviews in sequence, making auto-decisions at each gate using six consistent principles. Outputs a single decision log.

Spins up a headless browser to click through flows, fill forms, and check that pages load and behave as expected — without writing test code.

Fetches the current front page with titles, points, and comment counts. Useful for a quick read on what's trending in tech without the site visit.

Intercepts destructive commands—drops, deletes, resets—and requires explicit confirmation with a summary of what will be lost before executing.

Freezes the current conversation thread and gives you a link to resume it later without losing any work, messages, or Claude's understanding of what you were doing.

Points at a feature or module and auto-generates a structured README, API reference, or architecture guide by reading the code and filling in the blanks.

Takes error logs, stack traces, or a symptom description and walks backward through the dependency chain to pinpoint the real failure point, not just the symptom.

Enforces safety guardrails on file edits: warns before destructive commands, scopes changes to a directory, and requires confirmation for risky operations.

Connects to a physical iPhone or simulator, executes tap and assertion sequences on a SwiftUI app, and reports pass/fail with screenshots of failures.

Generates a dashboard showing test coverage, debt hotspots, dependency freshness, and build stability across your repo — metrics that matter for velocity.

Organizes scattered retrospectives, postmortems, and notes into a searchable knowledge base so the next project doesn't repeat yesterday's mistakes.

Executes your test suite and produces a detailed report of pass/fail results, without making changes to your codebase or CI/CD pipeline.

Reviews your staged changes line-by-line for logic errors, missed edge cases, and style drift — before you push to main.

Removes a freeze lock that was placed on protected directories, letting the team edit everywhere again after a release or hotfix window closes.

Reviews a quarterly roadmap or OKR plan for holes in logic, unspoken dependencies, and resource mismatches. Asks the hard questions a CEO-level reader would catch.

Methodically tests your app against a checklist—forms, navigation, edge cases, accessibility—and surfaces bugs with repro steps and severity. Works with live URLs.

Takes a loose problem statement and runs it through five clarifying phases — context, constraints, acceptance criteria, edge cases, and rollout — to produce a spec an engineer can build from.

Analyzes patterns in how developers answer questions and adjusts sensitivity and framing so follow-ups match their communication style and seniority, not a template.

Merges your branch, runs tests, bumps the version number, updates the changelog, and opens a PR—all in one command. Handles the tedious parts of shipping.

Runs a baseline benchmark, then re-runs it on your changes and flags any measurable slowdown with the exact diff that caused it.

Runs a skill through Claude 3.5 Sonnet, Opus, and Haiku side-by-side to see which model handles the prompt best, with latency and cost trade-offs.

Restores a saved project context (code, docs, state) from an earlier checkpoint, so you can pick up a multi-turn task days or weeks later without re-explaining.

Takes your release notes or changelog and updates the relevant doc pages, API reference, and examples to stay in sync without manual hunting through files.

Tests your API endpoints and SDK against real-world use patterns, surfacing the rough edges—unclear error messages, missing docs, tedious setup flows—that slow down early adopters.

Learns your product's brand and market, then generates a full design system—colors, typography, spacing, motion rules—with visual previews so you can iterate before handing to your team.

Takes a design spec or Figma screenshot and generates clean, responsive Pretext-native HTML/CSS ready to hand to engineering without rework.

A CLI wrapper that lets you invoke Codex in three modes—completion, search, and translation—without leaving your terminal or IDE.

Points Claude at a crash log or test failure in your iOS codebase. Claude reads the stack, finds the culprit file, and produces a tested patch you can merge.

Captures screenshots from actual iPhones, reviews spacing, typography, color contrast, and component consistency against Apple's Human Interface Guidelines, with fixes flagged per screen.

Evaluates your infrastructure, access controls, and incident response plan as a seasoned security leader would—surfaces gaps and rates each by business impact.

Scans a design file or screenshot for spacing drift, color bleed, broken hierarchy, and AI-generated artifacts. Flags each issue with coordinates and suggests a specific fix.

Produces 4–6 distinct design variants from a brief, opens them side-by-side, collects structured feedback from stakeholders, and tracks which direction won.

Two modes: simulate a YC partner's direct feedback on your deck or narrative, or drill into a specific problem area (unit economics, go-to-market, fundraising strategy) with the rigor of a real session.

Checks your current gstack version, fetches the latest release, and handles the upgrade without breaking existing configurations or data.

Prevents Claude from accidentally modifying files outside a safe directory during a session. Useful when working on a specific project or module in a larger codebase.

Turn any markdown file into a publication-quality PDF. (gstack)

Automates the handoff from code review to merge to deployment, running tests and rollback checks at each gate so broken code never reaches users.

Displays a read-only dashboard of what's queued, in progress, and shipped across your workspace. See blockers and timing at a glance without digging through Slack.

Regenerates your iOS debug bridge code against the current upstream gstack templates, ensuring compatibility with the latest framework changes and avoiding manual template drift.

Removes the DebugBridge package and all #if DEBUG conditionals from your codebase, leaving a lean production binary without manual file-by-file hunting.

Takes a roadmap, quarterly plan, or proposal and stress-tests it—questioning scope, identifying missing dependencies, calling out risky assumptions, and suggesting bolder or leaner versions.

Spins up a Chromium instance that Claude can navigate, click, and read — with the sidebar extension installed so you can talk to Claude about what's on screen.

Interactive developer experience plan review. (gstack)

Scans commits, PRs, and code quality from the past week, then generates a team retro with per-person wins, blockers, and growth areas—tracked over time to spot trends.

Sets up the credentials, environment variables, and infrastructure settings needed to deploy code to staging or production without manual config.

Prompts the team to answer what went well, what didn't, and what to change next week. Synthesizes responses into a readable summary and one concrete action.

Paste a stack trace, error log, or description of unexpected behavior. Extracts the failure signal, narrows the root cause, and suggests a fix with confidence levels for each hypothesis.

Takes a successful one-off scrape workflow and packages it into a permanent, shareable skill you can run again without rebuilding the logic.