GStack Skill Product Teardown

Product Frame

gstack turns an AI coding session into a structured software organization. The core product is not a CLI. It is a repeatable sprint operating system made of roles, gates, and tools:

Think -> Plan -> Build -> Review -> Test -> Ship -> Reflect.

The README makes the promise explicit: gstack "turns Claude Code into a virtual engineering team" with CEO, engineering manager, designer, reviewer, QA lead, security officer, and release engineer roles. The local repo currently exposes 52 SKILL.md files across core gstack, browser skills, OpenClaw-native skills, and host adapters.

Primary Personas

Persona	Real Job	Why gstack fits
Technical founder/CEO	Ship faster without losing product judgment	Encodes founder review, scope challenge, QA, and release discipline
Staff engineer/tech lead	Keep AI-generated work from becoming chaos	Adds architecture review, code review, tests, safety, and release gates
First-time Claude Code user	Avoid blank-prompt drift	Provides named workflows with clear triggers and outputs
Multi-agent operator	Run many AI workstreams in parallel	Gives each sprint a lifecycle and stop condition
Design/product builder	Make AI UI work inspectable and shippable	Adds design exploration, design review, HTML production, and browser QA

JTBD Map

Job	Trigger	gstack Module Family	Desired Outcome
Reframe a vague product idea	"I want to build X"	office-hours, plan-ceo-review, autoplan	Better problem definition before code
Turn a plan into an engineering plan	Feature idea is approved	plan-eng-review, plan-design-review, plan-devex-review	Architecture, test matrix, UX/DX risks surfaced
Implement without drift	Scope is approved	ship, careful, freeze, guard	Code stays bounded, tested, and reviewable
Catch AI production bugs	Branch has changes	review, codex, cso	Independent review, security review, model disagreement
Inspect the actual UI	App has URL or staging environment	browse, open-gstack-browser, qa, qa-only	Browser-level evidence instead of text-only claims
Release safely	PR is ready or approved	ship, land-and-deploy, canary, benchmark	Tests, PR, deploy verification, monitoring
Preserve learning	Session has decisions or fixes	learn, context-save, context-restore, sync-gbrain	Reusable memory and recoverable work context

Skill Families

1. Product and Strategy

Core skills: office-hours, plan-ceo-review, gstack-openclaw-office-hours, gstack-openclaw-ceo-review, retro, gstack-openclaw-retro.

Product job: force founder-quality judgment before implementation. The key pattern is not "generate requirements." It is adversarial product review: challenge premises, choose scope mode, and identify the 10-star version hiding inside a narrower request.

Requirements:

Ask forcing questions before implementation.
Preserve the user's actual pain language.
Output a design doc that downstream skills can read.
Distinguish expansion, selective expansion, hold-scope, and reduction modes.

2. Planning and Review Pipeline

Core skills: autoplan, plan-eng-review, plan-design-review, plan-devex-review, design-consultation.

Product job: turn a fuzzy feature into an implementation-ready plan with taste, architecture, and developer-experience checks.

Requirements:

CEO review decides whether the idea should expand or shrink.
Design review rates dimensions and asks for taste decisions.
Engineering review draws data flow, failure modes, state machines, and test matrix.
DX review evaluates time-to-hello-world, docs, onboarding, and magical moment.
Autoplan routes the reviews so the user sees only decisions that matter.

3. Build, Ship, and Release

Core skills: ship, land-and-deploy, setup-deploy, canary, benchmark, benchmark-models, landing-report.

Product job: convert work into a verified PR or production release.

Requirements:

Sync main before shipping.
Run tests and coverage audit.
Bootstrap test framework if absent.
Open a PR with useful context.
Merge only after approval/CI.
Verify production health after deploy.

4. Browser, QA, and Design Execution

Core skills: browse, open-gstack-browser, setup-browser-cookies, qa, qa-only, design-review, design-shotgun, design-html, pair-agent.

Product job: give agents eyes and let them inspect actual software rather than reason from code only.

Requirements:

Launch controlled Chromium sessions.
Import cookies when authenticated testing is needed.
Produce screenshots and real click-path evidence.
Generate, compare, and iterate design variants.
Turn approved mockups into production HTML with responsive layout.
Let other agents share the browser through scoped, isolated tabs.

5. Safety, Security, and Scope Control

Core skills: careful, freeze, guard, unfreeze, cso, investigate, canary.

Product job: prevent AI agents from over-editing, guessing, or shipping insecure work.

Requirements:

Warn before destructive commands.
Restrict edits to a declared path during risky work.
Investigate root cause before fixing.
Run OWASP + STRIDE threat review.
Use confidence gates and false-positive exclusions for security findings.
Monitor post-deploy behavior.

6. Memory and Context Continuity

Core skills: learn, context-save, context-restore, setup-gbrain, sync-gbrain.

Product job: make sessions compound. gstack captures project preferences and routes durable knowledge into gbrain when available.

Requirements:

Store project-specific learnings, decisions, and preferences.
Restore session context from checkpoint commits.
Register repo sources with gbrain.
Add or remove brain guidance based on capability checks.
Respect per-remote trust tiers: read-write, read-only, deny.

7. Documentation and Knowledge Products

Core skills: document-release, document-generate, make-pdf, scrape, skillify.

Product job: keep the written surface aligned with shipped reality.

Requirements:

Diff-aware doc updates.
Diataxis map: reference, how-to, tutorial, explanation.
Generate missing docs from codebase research.
Convert working docs into PDFs.
Turn external examples into reusable skills.

System Requirement Summary

gstack works because every skill has a narrow operating contract:

Trigger: when to invoke it.
Role: which specialist it simulates.
Inputs: plan, branch, URL, repo, design, or question.
Gates: tests, screenshot, decision, approval, or deploy health.
Output: artifact the next skill can consume.

The deeper product insight: gstack is a management system for AI labor. The names are slash commands, but the thing being sold is operational confidence.

Complete Skill Catalog

This catalog comes from reading the local gstack/README.md and all 52 gstack/**/SKILL.md files. The useful product lens is: each skill behaves like a named staff function with a trigger, expected inputs, output artifact, quality gate, and JTBD.

Core / Runtime

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`gstack`	`gstack/SKILL.md`	Runtime/router + browse docs	browsing, screenshots, inspect page	URL, page/task	browser actions, screenshots, QA evidence	update check, telemetry/privacy prompts, routing setup, browse build check	Turn agent into an instrumented operator, not a chat box.
`context-save`	`gstack/context-save/SKILL.md`	Session archivist	save progress/state	git state, decisions, remaining work	resumable context artifact	captures state before handoff	Prevent lost work across sessions.
`context-restore`	`gstack/context-restore/SKILL.md`	Session restorer	resume / where was I	saved context	working summary + next steps	chooses most relevant saved state	Make discontinuous work feel continuous.
`learn`	`gstack/learn/SKILL.md`	Memory librarian	show/prune/export learnings	project learnings	reviewed/searchable/pruned memory	asks before destructive pruning	Make project knowledge compound.
`plan-tune`	`gstack/plan-tune/SKILL.md`	Preference tuner	too many questions / show profile	question history, user overrides	question preferences, profile view	preference confirmation	Reduce friction by learning when to ask.
`health`	`gstack/health/SKILL.md`	Code quality analyst	health check / quality score	repo tools	0-10 quality dashboard, trend	type/lint/test/dead-code checks	Give engineering health a single visible dashboard.
`landing-report`	`gstack/landing-report/SKILL.md`	Release queue reader	version queue / open PR slots	PRs, sibling workspaces, VERSION	read-only queue snapshot	no mutations	Avoid version collisions before shipping.
`gstack-upgrade`	`gstack/gstack-upgrade/SKILL.md`	Self-updater	upgrade/update gstack	install mode	upgraded global/vendored gstack, changelog	detects install type, asks on choices	Keep the toolchain current without manual spelunking.

Think / Plan

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`office-hours`	`gstack/office-hours/SKILL.md`	YC Office Hours	brainstorm, idea, worth building	product idea/context	design doc	six forcing questions; startup vs builder mode	Reframe feature asks into real demand and wedge.
`plan-ceo-review`	`gstack/plan-ceo-review/SKILL.md`	CEO/founder	think bigger, strategy review	plan/design doc	scope decision, premise challenges, revised plan	interactive mode choice	Find the 10-star product hiding inside scope.
`plan-eng-review`	`gstack/plan-eng-review/SKILL.md`	Eng manager	architecture / implementation plan	plan, repo context	architecture notes, diagrams, test plan	interactive issue walkthrough	Force execution risk into the open before coding.
`plan-design-review`	`gstack/plan-design-review/SKILL.md`	Senior designer	UX/design plan review	UI plan	0-10 design score, improved plan	interactive design choices	Catch AI slop before implementation.
`plan-devex-review`	`gstack/plan-devex-review/SKILL.md`	DX lead	API/DX/onboarding review	developer-facing plan	persona analysis, TTHW benchmark, friction map	DX expansion/polish/triage modes	Make developer products feel magical, not merely functional.
`autoplan`	`gstack/autoplan/SKILL.md`	Review pipeline	run all reviews / autoplan	plan file	fully reviewed plan, decision audit, test artifact	strict CEO -> Design -> Eng -> DX order, final approval gate	Compress a review gauntlet into one supervised command.
`design-consultation`	`gstack/design-consultation/SKILL.md`	Design partner	design system / brand guidelines	product + landscape	`DESIGN.md`, font/color previews	researches existing design first	Create a durable design source of truth.

Build / Debug / Review

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`investigate`	`gstack/investigate/SKILL.md`	Debugger	debug, root cause, broken behavior	error, repo, logs	root cause, hypothesis, fix	iron law: no fix without cause; freeze hooks	Stop patching symptoms.
`review`	`gstack/review/SKILL.md`	Staff engineer	code review / pre-landing	diff vs base	findings, auto-fixes, review log	structural issue checks, asks for risky fixes	Find production bugs CI misses.
`codex`	`gstack/codex/SKILL.md`	Second opinion	Codex review/challenge/consult	diff or question	pass/fail review, adversarial critique, consult notes	Codex availability, mode selection	Add independent model judgment.
`cso`	`gstack/cso/SKILL.md`	Chief Security Officer	security audit / threat model	repo, infra, deps, CI	verified findings, exploit scenarios, trend log	confidence gate and false-positive exclusions	Make security high-signal enough to use regularly.
`careful`	`gstack/careful/SKILL.md`	Safety guard	be careful / prod mode	bash commands	destructive-command warnings	PreToolUse Bash hook	Prevent irreversible mistakes.
`freeze`	`gstack/freeze/SKILL.md`	Edit scope lock	only edit folder / freeze	allowed directory	session edit boundary	PreToolUse Edit/Write hook	Keep debugging contained.
`guard`	`gstack/guard/SKILL.md`	Full safety mode	lock it down / maximum safety	directory + shell commands	careful + freeze active	Bash/Edit/Write hooks	Combine command and file safety.
`unfreeze`	`gstack/unfreeze/SKILL.md`	Scope unlocker	unfreeze / allow all edits	current freeze state	cleared boundary	state-file check	Restore normal work after constrained mode.

Browser / QA / Performance

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`browse`	`gstack/browse/SKILL.md`	QA browser operator	open site, screenshot, dogfood	URL, flow, selectors	screenshots, snapshots, assertions	browser daemon/build check	Give the agent eyes and hands.
`open-gstack-browser`	`gstack/open-gstack-browser/SKILL.md`	Visible browser launcher	open browser / Chrome / side panel	browser request	Chromium with sidebar	launch/build checks	Let user watch and supervise browser work.
`setup-browser-cookies`	`gstack/setup-browser-cookies/SKILL.md`	Session manager	import cookies / login	real browser cookies/domain	authenticated headless session	interactive picker	Test logged-in flows without fake auth.
`pair-agent`	`gstack/pair-agent/SKILL.md`	Multi-agent coordinator	share browser / connect agent	remote agent	setup key, scoped tab access	scoped tokens, tab isolation	Let other agents operate in the same browser safely.
`qa`	`gstack/qa/SKILL.md`	QA lead	QA / test and fix	app URL, tier	bug fixes, commits, before/after evidence	quick/standard/exhaustive tiers, reverify each fix	Convert testing into fix-verified readiness.
`qa-only`	`gstack/qa-only/SKILL.md`	QA reporter	report only / do not fix	app URL	structured bug report, screenshots, repro steps	no code changes	Get evidence without mutating code.
`design-review`	`gstack/design-review/SKILL.md`	Designer who codes	visual QA / polish	live site/source	atomic visual fixes, before/after screenshots	screenshot verification, commit per fix	Make shipped UI look intentional.
`devex-review`	`gstack/devex-review/SKILL.md`	DX tester	test onboarding / DX audit	docs, CLI/API flow	DX scorecard, screenshots, TTHW	actually runs onboarding; compares plan vs reality	Measure developer experience by doing it.
`benchmark`	`gstack/benchmark/SKILL.md`	Performance engineer	page speed / web vitals	URL/pages	baseline/regression report	browser, resource, and vitals checks	Catch performance regressions before merge.
`canary`	`gstack/canary/SKILL.md`	SRE	post-deploy monitor	production URL	monitoring screenshots, anomaly alerts	console/perf/page failure loop	Verify production after deploy.
`scrape`	`gstack/scrape/SKILL.md`	Read-only extractor	scrape / pull data	page + extraction intent	JSON data	read-only boundary; routes to browser-skill if known	Turn pages into structured data.
`skillify`	`gstack/skillify/SKILL.md`	Skill codifier	save/codify scrape	successful scrape transcript	browser-skill dir, `script.ts`, test, fixture	provenance guard, temp-dir test, approval gate	Convert repeated browser work into durable automation.
`hackernews-frontpage`	`gstack/browser-skills/hackernews-frontpage/SKILL.md`	HN scraper	scrape HN top stories	HN front page	titles, points, comments	trusted host/script	Example of a codified browser skill.

Design / Docs / Artifacts

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`design-shotgun`	`gstack/design-shotgun/SKILL.md`	Design explorer	show options / variants	UI idea, taste memory	4-6 variants, comparison board, feedback	user approval/iteration	Explore taste space quickly.
`design-html`	`gstack/design-html/SKILL.md`	Design engineer	turn design into HTML	approved mockup/plan/description	production HTML/CSS	framework detection, Pretext pattern routing	Make the selected design real.
`document-generate`	`gstack/document-generate/SKILL.md`	Documentation author	write docs/tutorial/how-to	feature/module/project	Diataxis docs	codebase research before writing	Fill missing docs with structure.
`document-release`	`gstack/document-release/SKILL.md`	Post-ship tech writer	sync docs after ship	diff + existing docs	README/docs/changelog/TODOS updates, coverage map	drift detection, Diataxis coverage	Keep docs aligned with shipped reality.
`make-pdf`	`gstack/make-pdf/SKILL.md`	Document publisher	markdown to PDF	markdown file	polished PDF	render/text extraction checks	Turn Markdown into finished collateral.

Ship / Deploy

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`setup-deploy`	`gstack/setup-deploy/SKILL.md`	Deploy configurator	setup deployment	repo/platform	deploy config in `CLAUDE.md`	platform/URL/health detection	Make future deploys one-command.
`ship`	`gstack/ship/SKILL.md`	Release engineer	ship/create PR/deploy	branch diff	tests, review, version bump, changelog, PR	base sync, tests, review, verification, PR title rules	Convert local work into a mergeable PR.
`land-and-deploy`	`gstack/land-and-deploy/SKILL.md`	Release engineer	land PR / production deploy	approved PR	merged PR, CI/deploy verification	waits for CI/deploy, canary health	Finish the loop from approved to live.

GBrain / Knowledge Infrastructure

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`setup-gbrain`	`gstack/setup-gbrain/SKILL.md`	Brain onboarding	setup/connect gbrain	local/Supabase choice	CLI, MCP registration, trust policy	trust triad, engine/config checks	Give the agent searchable repo memory.
`sync-gbrain`	`gstack/sync-gbrain/SKILL.md`	Brain indexer	sync/reindex gbrain	repo code	refreshed index + `CLAUDE.md` search guidance	policy deny stop, local engine preflight, health verdict	Keep brain search current and trustworthy.
`benchmark-models`	`gstack/benchmark-models/SKILL.md`	Model evaluator	compare models / model shootout	prompt/skill	latency, tokens, cost, quality comparison	auth/provider availability, optional judge	Pick models by evidence, not vibes.

OpenClaw Native

Skill	Path	Role	Trigger / When Used	Inputs	Outputs / Artifacts	Gates / Checks	JTBD
`gstack-openclaw-office-hours`	`gstack/openclaw/skills/gstack-openclaw-office-hours/SKILL.md`	OpenClaw product coach	brainstorm / idea review	product idea	conversational office-hours review	conversational only	Bring gstack methodology into OpenClaw directly.
`gstack-openclaw-ceo-review`	`gstack/openclaw/skills/gstack-openclaw-ceo-review/SKILL.md`	OpenClaw CEO reviewer	challenge/review plan	proposal/plan	strategic critique	conversational only	Plan challenge without spawning Claude Code.
`gstack-openclaw-investigate`	`gstack/openclaw/skills/gstack-openclaw-investigate/SKILL.md`	OpenClaw debugger	debug/root cause	error/context	investigation path	root-cause framing	Apply gstack debugging method in chat.
`gstack-openclaw-retro`	`gstack/openclaw/skills/gstack-openclaw-retro/SKILL.md`	OpenClaw retro lead	weekly retro / what shipped	commits/history	team-aware retro	trend/history tracking	Turn engineering activity into learning.

Deep Product Insights

gstack is a role-based operating system for AI-assisted shipping, not a tool bundle. The README's "virtual engineering team" framing is structurally accurate.
The core product move is gates, not generation. Most skills add stop points, confidence thresholds, verification loops, approval moments, or hooks around otherwise cheap AI action.
The system optimizes for complete workflow ownership: idea -> reviewed plan -> implementation -> review -> QA -> docs -> PR -> deploy -> canary -> retro.
Browser control is foundational. QA, DX audit, design review, scraping, canary, benchmark, and pair-agent all depend on real page access and evidence capture.
The long-term bet is compounding memory: learnings, context save/restore, GBrain sync, taste memory, retros, and question tuning all make repeated work cheaper and more personalized.