SDLC Agent-Skill Wiki

gstack

10 min read

gstack

Workflow — the primary skill per SDLC stage this framework runs, top to bottom (folded and off-stage steps omitted). Validate and Review are sibling gates that both run after Implement.

flowchart TB
    A["gstack-office-hours<br/><i>Align</i>"] --> S["gstack-spec<br/><i>Specify</i>"] --> P["gstack-autoplan<br/><i>Plan</i>"] --> I["gstack-design-html<br/><i>Implement</i>"]
    I --> V["gstack-qa<br/><i>Validate</i>"]
    I --> R["gstack-review<br/><i>Review</i>"]
    V --> Rel["gstack-ship<br/><i>Release</i>"]
    R --> Rel
    Rel --> L["gstack-learn<br/><i>Learn</i>"]

gstack is Garry Tan’s (President & CEO of Y Combinator) open-source, MIT-licensed toolkit that turns Claude Code into a virtual engineering team — “a CEO who rethinks the product, an eng manager who locks architecture, a designer who catches AI slop, a reviewer who finds production bugs, a QA lead who opens a real browser, a security officer who runs OWASP + STRIDE audits, and a release engineer who ships the PR.” It is the largest framework in this wiki: ~54 skills (all slash-commands, all Markdown) plus standalone CLIs, organized around a full sprint:

Think → Plan → Build → Review → Test → Ship → Reflect

“gstack is a process, not a collection of tools. The skills run in the order a sprint runs” — each skill feeds outputs into the next (/office-hours writes a design doc /plan-ceo-review reads; /plan-eng-review writes a test plan /qa picks up; /review catches bugs /ship verifies). It is built for founders/CTOs/tech-leads who still ship, is engineered for 10–15 parallel sprints (via Conductor), and works across 10 coding agents (Claude Code, Codex, Cursor, Factory, …). Install: git clone … ~/.claude/skills/gstack && ./setup.

  • Root router: gstack-router — the gstack skill dispatches work to the right specialist skill (its counterpart is Addy’s addy-using-agent-skills).
  • Builder ethos (injected into every workflow preamble): Boil the Ocean (completeness is near-free with AI — do the complete thing), Search Before Building (three layers of knowledge; the “eureka” is zigging where others zag → pattern-source-grounding), User Sovereignty (models recommend, users decide — the generation-verification loop, Karpathy’s “Iron Man suit”).

Distinctive contributions to the ontology

  • The clean Review ∥ Test partition — it promoted a new canonical stage. gstack’s sprint separates a Review phase (code/design/security review: gstack-review, gstack-codex, gstack-design-review, gstack-devex-review) from a distinct Test phase (functional browser QA: gstack-qa, gstack-qa-only). As the second framework (after Addy) with that clean functional-validation ∥ quality-review partition, gstack cleared the bar to split stage-review out of stage-validate (2026-07-05): its Review-phase skills implement stage-review, its Test-phase skills stage-validate.
  • A genuine deploy + operate arc. gstack is the first framework here to separate deploy from ship: gstack-ship (PR) → gstack-land-and-deploy (merge, deploy, verify prod health) → gstack-canary (post-deploy monitoring). This strengthens the stage-release stage-operate split candidate.
  • A third framework for stage-learn. gstack-retro (team retrospective) + gstack-learn (a compounding cross-session learnings corpus) + gstack-skillify (codify a successful run into a permanent skill) give the Reflect stage a third framework alongside Compound Engineering and BMAD — beyond doubt now.
  • A third framework for pattern-persona-agents. gstack is thoroughly persona-oriented (CEO / Eng Manager / Senior Designer / DX Lead / Staff Engineer / QA Lead / CSO / SRE / Release Engineer), joining BMAD as the second framework whose whole toolkit is named expert personas.

Signature mechanisms

Capabilities by sprint stage

Think — stage-align

  • gstack-office-hours — YC Office Hours; reframe the product with six forcing questions before any code; writes the design doc every downstream skill reads.

Plan — stage-plan

  • gstack-plan-ceo-review — CEO/founder review; find the 10-star product; four scope modes.
  • gstack-plan-eng-review — Eng-manager review; lock architecture, data flow, edge cases, tests.
  • gstack-plan-design-review — Senior-designer review; rate each design dimension 0-10; AI-slop detection.
  • gstack-plan-devex-review — Developer-experience review; TTHW, personas, magical moment, friction traces.
  • gstack-autoplan — run CEO → design → eng → DX reviews sequentially with auto-decisions; surface only taste calls.
  • gstack-plan-tune — self-tune AskUserQuestion sensitivity + developer psychographic (plan-review meta).
  • gstack-design-consultation — build a complete design system from scratch (aesthetic/type/color/layout/motion) → DESIGN.md.

Specify — stage-specify

  • gstack-spec — turn vague intent into a precise, executable spec in five phases; Codex quality gate; files a GitHub issue; --execute spawns a fresh-worktree agent.

Build — stage-implement

Review — quality gate (stage-review)

  • gstack-review — Staff-Engineer pre-landing PR review; finds bugs that pass CI but break in prod; auto-fixes the obvious.
  • gstack-codex — OpenAI Codex second opinion; review / adversarial-challenge / consult modes; cross-model analysis.
  • gstack-design-review — Designer-who-codes live visual audit + fix loop with atomic commits.
  • gstack-devex-review — live developer-experience audit; measures TTHW against the real flow (the /plan-devex-review boomerang).
  • gstack-cso — Chief Security Officer; OWASP Top 10 + STRIDE threat model with concrete exploit scenarios.
  • gstack-health — code-quality dashboard (type checker, linter, tests, dead code).

Test — functional QA (stage-validate)

  • gstack-qa — QA Lead; real browser, find bugs, fix with atomic commits, re-verify, auto-generate regression tests.
  • gstack-qa-only — same methodology, report-only (no code changes).
  • gstack-benchmark — performance regression: page load, Core Web Vitals, resource sizes, before/after per PR.

Ship — stage-release

  • gstack-ship — Release Engineer; sync base, run tests, review diff, bump VERSION, update CHANGELOG, commit, push, open PR.
  • gstack-land-and-deploy — merge the PR, wait for CI + deploy, verify production health.
  • gstack-canary — SRE; post-deploy monitoring loop (console errors, perf regressions, page failures).
  • gstack-landing-report — read-only dashboard for the workspace-aware ship queue.
  • gstack-document-release — post-ship doc update; Diataxis coverage map in the PR body.
  • gstack-document-generate — generate missing Diataxis docs (tutorial/how-to/reference/explanation) from code.
  • gstack-setup-deploy — one-time deploy-config detection (Fly.io / Render / Vercel / …).
  • gstack-ios-clean — strip the DebugBridge + #if DEBUG wiring before a Release build.

Reflect — stage-learn

  • gstack-retro — team-aware weekly retrospective; per-person breakdowns, shipping streaks, test-health trends.
  • gstack-learn — manage the cross-session learnings corpus (patterns/pitfalls/preferences) so gstack compounds on your codebase.
  • gstack-skillify — codify the most recent successful gstack-scrape into a permanent browser-skill.

iOS live-device QA

Browser & agent-integration infrastructure

Memory & knowledge infrastructure

  • gstack-setup-gbrain — set up gbrain persistent cross-machine memory (PGLite / Supabase / remote MCP).
  • gstack-sync-gbrain — re-index this repo’s code into gbrain and refresh CLAUDE.md search guidance.

Cross-cutting utilities (no single lifecycle stage)

  • gstack-diagram — English (or mermaid) → diagram triplet (mermaid + editable .excalidraw + SVG/PNG), offline.
  • gstack-make-pdf — any markdown → publication-quality PDF (renders mermaid/excalidraw fences), --to html|docx.
  • gstack-benchmark-models — cross-model benchmark for gstack’s own skills (Claude / GPT / Gemini) (meta/tooling).
  • gstack-upgrade — self-updater; detects global vs vendored install (meta/tooling).

Artifacts produced

Patterns applied

See Also

Graph View

Backlinks