spec

An agent skill by garrytan, from garrytan/gstack. Tags: documentation, planning, spec, validation.

What it does

Turn vague intent into a precise, executable spec in five phases. (gstack)

Install

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

npx skills add garrytan/gstack --skill spec

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/spec ~/.claude/skills/spec

Safety box score

Grade F, high risk: 70 of 100 risk points, rated 2026-10-20. 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 turns a request into a GitHub issue, archives the spec locally, and can optionally spin up a second `claude` process to execute it. The main risks are outbound data leaving the machine, writes into home-dotdir and git metadata, and a runtime instruction system that can alter agent behavior based on tool output.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	3 (high)	It shells out to local helpers and external CLIs throughout the flow, including `~/.claude/skills/gstack/bin/gstack-skill-start`, `bun`, `gh`, `codex exec`, and `claude -p`, so the skill directly runs commands and spawned processes.
Network access	2 (moderate)	It sends spec content to remote services via `gstack-post issue-create`, `gh issue list/view`, and the Codex outside-review path, but the outbound flow is bounded to the spec and review artifacts.
Filesystem reach	2 (moderate)	It writes outside the project into `~/.gstack/`, `~/.claude/skills/gstack/`, and `.git/info/exclude`, and it also creates and deletes temp work files and worktrees.
Secrets & credentials	2 (moderate)	The redaction pipeline inspects the exact bytes before upload and passes `git config user.email` into the scanner, and the workflow depends on existing `gh`/Codex auth state, so it touches auth-adjacent data.
Privilege & persistence	2 (moderate)	It changes local git settings and permissions (`.git/info/exclude`, `chmod 700`), but it does not require sudo or install system services.
Instruction hijack surface	2 (moderate)	It consumes `GSTACK_INSTRUCTION_BEGIN` blocks from tool output and explicitly says to treat the skill file as executable instructions; spawned sessions also auto-choose options instead of asking.
Irreversible actions	3 (high)	Filing the issue and, in some modes, spawning a fresh agent/worktree are built into the workflow, so the skill can take durable external actions without a fresh human check at every step.
Transparency	1 (low)	Most logic is readable, but key steps are delegated to external helper binaries not included here (`gstack-redact`, `gstack-post`, `gstack-codex-probe`), so some behavior is outside the reviewed files.

Before installing

Verify `gh` and Codex auth are present on this machine, since the skill will use both if available.
Read `SKILL.md` and `sections/gate-and-file.md` and confirm you want writes under `~/.claude/`, `~/.gstack/`, and `.git/info/exclude`.
Test it in a scratch repo first if you do not want a real GitHub issue filed on the first run.

Rated 2026-10-20 by an automated review of 3 files from the skill's repository.

Source

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

marketing-plan — When the user needs a comprehensive marketing plan for a client, a company they advise, or their own product.
principle-sequence-verifiable-units — Apply to multi-step work (sweeps, migrations, runs of similar edits) and to how you stack commits and PRs.
seo-content — Content quality and E-E-A-T analysis with AI citation readiness assessment.
verify-this — Verify a claim with fresh local evidence: restate it falsifiably, capture baseline and treatment, compare artifacts, and return VERIFIED, NOT VERIFIED.
X MCP guide — ALWAYS read this when a user connects the X plugin or any X MCP, before using any X connection, and again on any X error.
claude-md-progressive-disclosurer — Optimize, slim, or restructure CLAUDE.md/AGENTS.md with progressive disclosure and zero information loss. Use when the user explicitly asks to audit, 精简, 瘦身.