Skills Explorer Add Skills

review

An agent skill by garrytan, from garrytan/gstack. Tags: code-review, deployment, review, validation.

What it does

Pre-landing PR review. (gstack)

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add garrytan/gstack --skill review

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/garrytan/gstack
cp -r gstack/review ~/.claude/skills/review

Safety box score

Grade F, high risk: 81 of 100 risk points, rated 2026-10-16. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill is a full review orchestrator: it runs shell commands, reads git/PR state, searches local history, and can post review replies and telemetry from the user's authenticated accounts. The main risks are broad local execution and home-directory writes, plus instruction-following from generated tool output and some review logic that lives in unreadable or external helper files.

Category levels, from 0 (none) to 3 (high)
CategoryLevelWhy
Code & shell execution3 (high)It directly runs shell helpers and repo commands such as `~/.claude/skills/gstack/bin/gstack-skill-start`, `git fetch`, `gh api`, `bun`, and `gstack-post`, so the agent can execute arbitrary local commands while reviewing.
Network access2 (moderate)It fetches PR data and comments with `gh api`, can use `WebSearch`/Aside, and sends telemetry plus GitHub replies via `gstack-post` and `gstack-skill-end`.
Filesystem reach3 (high)It reads and writes in `~/.claude/`, `~/.gstack/`, and `.gstack/tmp/`, and the plan/review flow explicitly writes logs, question files, telemetry, and review artifacts outside the project tree.
Secrets & credentials2 (moderate)It operates against authenticated `gh`/`glab` sessions and processes PR bodies, comments, and reply files that may contain sensitive material; the greptile workflow also warns about credential-like text in replies.
Privilege & persistence1 (low)It does not request sudo or install services, but it does change local user state such as gstack config and home-directory artifacts; the blast radius stays at the user's account level.
Instruction hijack surface3 (high)`SKILL.md` tells the agent to treat generated output blocks as directives and to auto-choose recommended options in spawned/unattended sessions, which is a direct instruction-following path from tool output into behavior.
Irreversible actions3 (high)It can post GitHub replies/comments, log telemetry, and auto-decide review dispositions without asking the user each time, so it can speak and act on the person's behalf.
Transparency3 (high)Two referenced files could not be read here (`checklist.md` and `sections/adversarial.md`), and most behavior is delegated to external helpers under `~/.claude/skills/gstack/bin`, so a reviewer cannot fully inspect the implementation from the shipped files alone.

Before installing

Rated 2026-10-16 by an automated review of 17 files from the skill's repository.

Source

Repository
garrytan/gstack (all skills from this repository)
Path
review/SKILL.md
Branch
main
Updated
2026-10-05