plan-devex-review

An agent skill by garrytan, from garrytan/gstack. Tags: developer-tools, planning, review, workflow.

What it does

Interactive developer experience 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-devex-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/plan-devex-review ~/.claude/skills/plan-devex-review

Safety box score

Grade F, high risk: 81 of 100 risk points, rated 2026-10-19. 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 full DX review runner, not a read-only doc helper. It shells out to many local commands, rewrites the plan/report, writes persistent logs under `~/.gstack/`, and can send the plan context to an external Codex reviewer and web search. The main risks are broad filesystem reach, runtime-instruction injection, and dependence on uninspected helper binaries.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	3 (high)	The skill is execution-heavy: it runs `bash`, `gstack-skill-start`, `gstack-skill-end`, `gstack-review-log`, `gstack-review-read`, and can invoke `codex exec`; that is inherent to its purpose but grants broad command execution.
Network access	3 (high)	It can research the web and, by default, ship the plan/context to an external Codex reviewer (`codex exec`) and web search; that is deliberate off-box data flow with a large blast radius.
Filesystem reach	3 (high)	The workflow writes review artifacts under `~/.gstack/`, `.gstack/tmp/`, and the plan file itself, and it deletes temp trees with `rm -rf`; that reaches outside the project workspace.
Secrets & credentials	1 (low)	It probes authenticated CLI state (`gh auth status`, `glab auth status`, `codex login`) but the visible files do not directly read token contents or secret files, so the secret surface is limited.
Privilege & persistence	2 (moderate)	It mutates persistent user-level configuration such as `gstack-config set cross_project_learnings true/false` and writes to home-directory state, but it does not require sudo or system services.
Instruction hijack surface	3 (high)	It tells the agent to treat the skill file as executable instructions, honor `GSTACK_INSTRUCTION_*` blocks from tool output, and auto-choose recommended options in spawned/unattended runs; that makes untrusted runtime text part of the control plane.
Irreversible actions	2 (moderate)	It can automatically rewrite the plan/report, log telemetry, and launch an outside review without another human decision, though it does not push code or pay money.
Transparency	3 (high)	Core behavior is delegated to helper binaries that are not included in the reviewed files (`gstack-skill-start`, `gstack-codex-probe`, `gstack-review-log`, `gstack-review-read`, `gstack-brain-cache`), so the real side effects are not fully auditable here.

Before installing

Read `sections/review-sections.md` and decide whether you want the default-on outside voice; it can send plan context off-box.
Inspect the helper binaries on your machine (`gstack-skill-start`, `gstack-review-log`, `gstack-codex-probe`); their behavior is not visible in these files.
Run it on a scratch branch first if you do not want the review loop rewriting your plan/report and writing under `~/.gstack/`.
If cross-project learning or CLI auth state matters to you, check those settings before enabling the skill.

Rated 2026-10-19 by an automated review of 5 files from the skill's repository.

Source

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

cavecrew — When to delegate to `cavecrew-investigator` (locate code).
cli-for-agents — Designs or reviews CLIs so coding agents can run them reliably: non-interactive flags, layered --help with examples, stdin/pipelines, fast actionable errors.
code-improver — Runs an autonomous review-and-fix improvement loop over any code target — a skill, plugin, module, or directory — using a reviewer the user names.
create-verification-skill — Generate a project-local verification skill that drives your app the way a user does — any language, framework, or platform.
events — When the user wants to plan, run, sponsor, speak at, or get pipeline from events — webinars, conferences, trade shows, meetups, dinners, workshops.
expo-overview — Framework (OSS). Entry point and router for every Expo or EAS task.