Harnesses > Workflow methodologies
gstack
Draft. This entry has not been checked against its sources yet.
Garry Tan's set of 55 skills, written for Claude Code, that runs a sprint through role-based steps, from product questioning and plan reviews to specialist code review, QA in a real browser, and a /ship command that ends with a pull request.
| Maintainer | Garry Tan (garrytan) |
|---|---|
| Category | Workflow methodologies |
| Hosts | Claude Code, Codex, Cursor, Factory Droid, Kiro, OpenCode |
| Components | 55 skills, 11 hooks |
| License | MIT |
| Version | 1.91.2.0 |
| Source code | https://github.com/garrytan/gstack |
| Website | none |
| Docs | https://github.com/garrytan/gstack/blob/main/docs/skills.md |
| DeepWiki | https://deepwiki.com/garrytan/gstack |
| Last researched | 2026-09-28, at commit 01593aa |
Features
Definitions are on the compare page.
| Feature | Supported | Notes |
|---|---|---|
| Stages of work it covers | ||
| Brainstorming | yes | /office-hours asks one question at a time, challenges your premises, and makes you pick one of two or three approaches before it writes a design doc. A hard gate forbids code. |
| Written plans | partial | You draft the plan in the host's plan mode; /plan-ceo-review, /plan-eng-review, and /autoplan review it. Design docs, review reports, and test plans go to ~/.gstack/projects/, not the repository. |
| Test first | no | Tests follow the code: /ship writes missing coverage tests, and /qa and /investigate add a regression test after each fix. The README's 'test-first execution' is not in the ship skill. |
| Debugging process | yes | /investigate: 'NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST', hypotheses tested before any fix, a stop after three failed hypotheses, and edits locked to the affected directory. |
| Code review | yes | /review sends specialist reviewer subagents, chosen by what the diff touches, plus an adversarial pass that goes to the Codex CLI when installed. /ship runs the same review before it pushes. |
| Verification | yes | /ship: 'NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE', checked against a ledger of test runs. /qa re-tests each fix in a browser. An opt-in Stop hook blocks the turn until a declared command passes. |
| Learning capture | yes | 48 of 55 skills end by logging learnings to ~/.gstack/projects/<slug>/learnings.jsonl, and 13 search it. Kept on your machine unless you turn on sync to a private git repository. |
| Branch to PR | yes | /ship: merge the base branch, test, review, bump VERSION, write CHANGELOG, split commits, push, and open the PR (GitHub or GitLab). /land-and-deploy merges and checks production (GitHub only). It does not create branches. |
| How it runs | ||
| Single entry point | yes | The gstack skill sends a request to the right skill. A 'Skill routing' section it offers to add to CLAUDE.md applies the same rules in every session. |
| Automatic activation | yes | The proactive setting is on by default and skills say 'Proactively invoke this skill'. On Claude Code descriptions are cut to one line, so matching leans on the CLAUDE.md routing section. |
| Session-start hook | no | The SessionStart hook, registered in team mode only, updates gstack in the background and injects nothing. Standing instructions go into CLAUDE.md. |
| Blocking hooks | yes | /careful blocks destructive commands, /freeze blocks edits outside one directory, /autoplan blocks skipping a phase, and an opt-in Stop hook holds the turn until a declared check passes. |
| Subagents | yes | |
| Parallel agents | yes | Specialist reviewers and design variants are launched in one message. Running many sessions at once ('10-15 parallel sprints') relies on the separate Conductor app. |
| Git worktrees | partial | Only /spec --execute creates a worktree, for a claude -p agent it spawns. The README leaves parallel sessions to Conductor workspaces. |
| Adopting it | ||
| Multiple hosts | yes | ./setup installs for six hosts. Claude Code is the source; the others get generated copies without hooks. OpenClaw, Hermes, and Slate get lighter integrations. |
| Per-project setup | yes | Team mode commits a CLAUDE.md section and, in required mode, a project hook that blocks skills until each developer installs gstack. Installs stay per user and pull the latest version; no version pin. |
| Team extensions | partial | Config keys, custom /careful warning patterns, /plan-tune question preferences, and /skillify browser scripts. No documented way to add workflow skills or change one without forking. |
| No extra services | yes | Needs Bun and builds a local Chromium. Reviews use the Codex CLI by default when it is installed and signed in; design mockups need an OpenAI API key. Telemetry is off unless you opt in. |
How it works
The workflow
The README calls gstack "a process, not a collection of tools" and orders the skills as a sprint: Think, Plan, Build, Review, Test, Ship, Reflect. Each step saves something the next one reads: "/office-hours writes a design doc that /plan-ceo-review reads. /plan-eng-review writes a test plan that /qa picks up."
- Think (
/office-hours). It asks what you are building for, then runs one of two modes. Startup mode asks six "forcing questions" about demand, the status quo, and the narrowest first version; builder mode brainstorms. It states premises for you to accept or reject, requires two or three approaches (one minimal, one ideal), and stops until you pick one. Then it writes a design doc to~/.gstack/projects/<slug>/. A hard gate forbids any code./specis the alternative when the result should be a GitHub issue. - Plan, in the host's plan mode.
/plan-ceo-reviewchallenges scope in one of four modes (expansion, selective expansion, hold scope, reduction)./plan-eng-reviewsettles architecture, data flow, edge cases, and a test plan./plan-design-reviewand/plan-devex-reviewcover interfaces and developer-facing work./autoplanruns them in sequence (design and developer-experience only when they apply, eng always last), answers routine questions with six written decision principles, and saves the judgment calls for one approval at the end. Each review appends a## GSTACK REVIEW REPORTto the plan file. - Build. gstack has no implementation skill. You approve the plan, leave plan mode, and the host writes the code.
- Review (
/review). It picks specialist reviewers from the diff: testing and maintainability always (for 50 or more changed lines), and security, performance, data migration, API contract, design, and simplification when the diff touches those areas, plus an adversarial reviewer. It fixes mechanical findings itself, asks about the rest, and repeats, up to three rounds, until a pass makes no edits./investigatehandles bugs,/csosecurity audits, and/design-reviewvisual problems. - Test (
/qa <url>). It drives a browser through the app, fixes each bug in its own commit, re-tests it, and adds a regression test./qa-onlyreports without fixing. - Ship (
/ship). From a feature branch it merges the base branch, runs the tests, audits coverage and writes missing tests, checks the plan's items were done, runs the pre-landing review, bumpsVERSION, writesCHANGELOG.md, updatesTODOS.md, splits the work into commits, checks the test evidence, pushes, updates docs through a/document-releasesubagent, and opens or updates the pull request./land-and-deploymerges, waits for CI and the deploy, and checks production./canarywatches it afterwards. - Reflect (
/retro). A weekly retrospective built from git history and gstack's logs./learnreviews and prunes saved learnings.
48 of the 55 skills share a generated preamble. It checks for updates and sets the question format and writing rules; in 39 skills it also lists recent artifacts and decisions for the current project and branch. These 48 skills end with a status (DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT) and a step that logs learnings.
Components
55 skills on Claude Code, each invoked as a slash command:
- router (1):
gstack - think and plan (7):
office-hours,spec,plan-ceo-review,plan-eng-review,plan-design-review,plan-devex-review,autoplan - design (5):
design-consultation,design-shotgun,design-html,design-review,diagram - review and audit (6):
review,deslop-shared-libs,codex,cso,devex-review,health - debugging (1):
investigate - browser and QA (9):
qa,qa-only,browse,scrape,skillify,open-gstack-browser,setup-browser-cookies,pair-agent,benchmark - ship and deploy (7):
ship,land-and-deploy,setup-deploy,canary,landing-report,document-release,document-generate - reflect and memory (5):
retro,learn,context-save,context-restore,plan-tune - safety (4):
careful,freeze,guard,unfreeze - iOS (5):
ios-qa,ios-fix,ios-design-review,ios-clean,ios-sync - utilities (5):
make-pdf,gstack-upgrade,setup-gbrain,sync-gbrain,benchmark-models
Other hosts also get
claude-code, which asks the Claude Code CLI for an outside review. The README's count, "Twenty-three specialists and eight power tools", is out of date.- router (1):
Skills are generated from
SKILL.md.tmpltemplates byscripts/gen-skill-docs.ts, which fills in shared blocks such as the preamble and the reviewer dispatch. The generated Claude Code skills total about 38,000 lines, plus 47 section files (about 13,600 lines) that skills read when a step needs them.No commands and no subagent definition files. Subagents are general-purpose agents given prompts from inside the skills.
agents/openai.yamlis Codex metadata.11 hooks, all for Claude Code:
- seven in
~/.claude/settings.json: a SessionStart auto-update (team mode only); three hooks on the question tool (a question log for/plan-tune, "never ask" preferences, and an error fallback), installed with your consent; a Stop hook, on by default, that closes the session timeline; the opt-ingstack-verify-gateStop hook; and an opt-in hook that passes prompts to the third-party Memorable CLI - three that load with a skill:
careful's destructive-command check,freeze's edit boundary (also used byguardandinvestigate), andautoplan's phase guard - one that team mode's
requiredsetting writes into your repository, which blocks skills until gstack is installed
- seven in
Tools:
browse, a headless-browser CLI (on macOS the browser skills drive the Aside browser first and fall back to it);design, which generates mockups with an OpenAI image model;make-pdf; a Chrome side-panel extension; and 89gstack-*scripts inbin/for settings, learnings, review logs, test evidence, telemetry, and hook management.No MCP servers of its own.
/setup-gbraincan register the server of GBrain, the author's separate knowledge-base project.
How it steers the agent
- Hard rules and stop points: "HARD GATE: Do NOT ... write any code" (
/office-hours), "NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST" (/investigate), "IRON LAW: NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE" (/ship), and explicit STOP points that wait for your answer. - Roles: the main skills cast the agent as a role (office-hours partner, CEO, eng manager, staff engineer, QA lead, release engineer), and the plan reviews list how that role thinks.
- Structured questions: questions go through the question tool as a "decision brief" with a plain explanation, a recommendation, and, where options differ in coverage, a "Completeness: X/10" score for each.
/plan-tunerecords your answers and, through a PreToolUse hook, can answer questions you have marked "never ask". - Evidence ledgers:
gstack-evidencerecords which test command ran against which working-tree content, and/shipaccepts only fresh evidence.gstack-review-logties each review pass to the content it saw. Shipping needs a clean eng review unless you setskip_eng_review. - Hooks that block:
/carefulasks before destructive commands and refuses recursive deletes of the root or home directory and force-pushes to the default branch./freezerefuses edits outside one directory./autoplanrefuses to load the next review phase until the current one is published. The opt-ingstack-verify-gatekeeps the turn open until your declared check passes. - A second model:
/review,/ship, the plan reviews, and/autoplanalso send the work to the Codex CLI when it is installed and signed in (to the Claude Code CLI when the host is Codex), and fall back to a subagent of the same model otherwise./office-hoursasks first. - Tests of the skills: 975 test files, including paid end-to-end evals that run skills through
claude -pand grade them with a model.
Files it writes
- On your machine, in
~/.gstack/projects/<slug>/: design docs, CEO plans, eng review reports and test plans, spec archives,/context-savecheckpoints,learnings.jsonl,timeline.jsonl, review logs, a decision log, and a design taste profile. 39 skills read recent items from here when they start, and plan reviews look up the latest design doc. Teammates do not see these files unless you turn on sync to a private git repository (gstack-artifacts-init) or use GBrain. - In the repository:
TODOS.md,CHANGELOG.md,VERSION, updated docs,DESIGN.md(from/design-consultation),.context/retros/(from/retro), report folders under.gstack/(QA, canary, benchmark, deploy), regression tests, and sections ofCLAUDE.md. - The host's plan file, to which plan reviews append their report.
~/.gstack/config.yamlfor settings, and~/.gstack/security/egress.jsonl, a hash-chained record of every network send gstack makes.
Install
gstack is not a Claude Code plugin. The README says to paste this into Claude Code:
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup
and to have Claude add a "gstack" section listing the skills to CLAUDE.md. ./setup builds the browser, links each skill into ~/.claude/skills/, and registers the hooks.
For a team, run ./setup --team, then ~/.claude/skills/gstack/bin/gstack-team-init required (or optional) inside the repository, and commit .claude/ and CLAUDE.md. Each developer still installs gstack; team mode then pulls the latest version at the start of each session, at most once an hour.
Other hosts: ./setup --host codex, opencode, cursor, factory (Factory Droid), or kiro, or let ./setup detect what is installed. These hosts get generated copies of the skills without hooks. OpenClaw and Hermes get prompt templates and generated skill files rather than an install; OpenClaw's main path is to start Claude Code sessions that have gstack, and four skills are published for it on ClawHub. Slate reads the Claude Code install. Any agent that reads a rules file can use the 2 KB summary in agents-digest/gstack-AGENTS.md.
Needs Git and Bun, plus Node.js on Windows. The Aside browser is recommended on macOS. /cso needs a native build toolchain. Remove it with bin/gstack-uninstall.
When to choose it
- You use Claude Code and want one command per stage: product questions, plan reviews, code review, browser QA, and a
/shipthat ends with a pull request. - You build web apps and want the agent to test them in a real browser and fix what it finds.
- You want some rules enforced by hooks rather than wording: a destructive-command check, an edit boundary, and an optional gate that holds the turn until tests pass.
- You run many sessions at once, for example in Conductor, and want each to follow the same stages.
When not to
- You want test-first development. gstack writes tests after the code.
- You want plans and lessons in the repository for the whole team to read. They stay in
~/.gstackon each machine. - You want a small instruction surface. The largest skills are 73 to 85 KB each, and the generated skills total about 38,000 lines.
- You want a stable process.
CHANGELOG.mdlists 277 versions between March and September 2026, and team mode moves everyone to the latest version automatically. - You do not want code sent to a second provider. With the Codex CLI installed and signed in, reviews send the diff to it by default; turn this off with
gstack-config set codex_reviews disabled. - Your main host is not Claude Code. Other hosts get translated skills without hooks, and several gates assume Claude Code's plan mode and question tool.
The first /office-hours session ends with a note from the author inviting you to apply to Y Combinator; later sessions skip it.
Sources
- garrytan/gstack at 01593aa:
README.md,ETHOS.md,CONTRIBUTING.md,CHANGELOG.md,docs/skills.md,docs/OPENCLAW.md,SKILL.md.tmpland*/SKILL.md.tmplwith theirsections/, the generated*/SKILL.md,scripts/gen-skill-docs.ts,scripts/resolvers/,hosts/,setup, andbin/(gstack-settings-hook,gstack-session-update,gstack-team-init,gstack-config,gstack-paths,gstack-verify-gate,gstack-skill-start) - DeepWiki was not used: it was unreachable from the research environment.