plan-eng-review
An agent skill by garrytan, from garrytan/gstack. Tags: architecture, engineering, planning, review.
What it does
Eng manager-mode plan review. (gstack)
Install
With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:
npx skills add garrytan/gstack --skill plan-eng-reviewOr 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/plan-eng-review ~/.claude/skills/plan-eng-reviewSafety box score
Grade D, elevated risk: 46 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 is a review-only skill, but it still runs a lot of local shell and helper scripts, writes review/state files under the user’s home directories, and can send plan content to remote services for outside review/search. The main risks are broad file access outside the repo, off-machine transmission during research, and auto-answer behavior in spawned/unattended modes that can reduce human checkpointing.
| Category | Level | Why |
|---|---|---|
| Code & shell execution | 2 (moderate) | It runs Bash plus many helper binaries, for example `~/.claude/skills/gstack/bin/gstack-skill-start --skill "plan-eng-review" --model "claude"`, so the skill has real local code-execution reach under the user account. |
| Network access | 2 (moderate) | It can send prompts and plan content off-machine via Aside/WebSearch and the outside-review path, including the `codex exec` prompt built from the current plan and decisions. |
| Filesystem reach | 2 (moderate) | It writes review artifacts and logs outside the repo, including `$GSTACK_STATE_ROOT/projects/$SLUG/...`, `~/.gstack/`, `.gstack/tmp/`, and the selected report file, so the blast radius reaches home-state directories. |
| Secrets & credentials | 0 (none) | The reviewed skill does not directly read password stores, browser sessions, keychains, or `.env` secrets; it only references repo metadata, branch names, and local config/state. |
| Privilege & persistence | 0 (none) | There is no sudo, global install, service registration, shell startup edit, or OS-setting change in the skill text. |
| Instruction hijack surface | 2 (moderate) | The skill tells spawned/unattended sessions to auto-choose recommended options and skip AskUserQuestion, and it contains prompt text that tells outside reviewers to ignore installed skills and fetched instructions. |
| Irreversible actions | 2 (moderate) | It can independently write review reports, logs, and decisions, and in spawned/unattended mode it makes decision choices without a human prompt, which broadens workflow autonomy even though it is not deploying or paying. |
| Transparency | 2 (moderate) | Most behavior lives in helper scripts under `~/.claude/skills/gstack/bin/` that are referenced but not included here, so a reviewer cannot fully inspect the actual side effects from the provided files alone. |
Before installing
- Read the helper scripts it calls, especially `gstack-skill-start`, `gstack-review-log`, `gstack-review-read`, and `gstack-codex-probe`.
- Decide whether you are okay with writes under `~/.gstack/` and `.gstack/tmp/`, not just inside the repo.
- If off-machine research/review is a problem, disable the Aside/Codex paths before use.
- Check whether auto-answer behavior in spawned/unattended sessions matches your review policy.
Rated 2026-10-16 by an automated review of 4 files from the skill's repository.
Source
- Repository
- garrytan/gstack (all skills from this repository)
- Path
- plan-eng-review/SKILL.md
- Branch
- main
- Updated
- 2026-10-05
Related skills
- architect — Sketch types, signatures, and module structure before code, then stay in the loop while implementation fills in.
- expo-project-structure — Framework (OSS). Folder structure for a new Expo app.
- principle-foundational-thinking — Apply before writing logic: choosing core types and data structures, sequencing scaffold-vs-feature work, asking what concurrent actors share.
- principle-outcome-oriented-execution — Apply during planned rewrites and migrations with explicit phase boundaries.
- run-learning-retrospective — Evaluate learning progress, identify blockers, and adjust the learning plan
- site-architecture — When the user wants to plan, map, or restructure their website's page hierarchy, navigation, URL structure, or internal linking.