gstack ships forty-seven skills. One top-level meta-skill at the repository root routes user requests to the right specialist. The remaining forty-six split into three groups. Twenty-three lifecycle specialists run a sprint end to end. Eight power tools sit alongside as safety guardrails, second opinions, and self-updaters. Fifteen supporting skills are infrastructure. Orchestration, semantic memory, context save and restore, browser primitives, and code-quality dashboards. Each card below names the slash command, the persona it role-plays inside the model, and the single action it owns.Documentation Index
Fetch the complete documentation index at: https://zikun.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
Think
/office-hours
YC Office Hours partner. Six forcing questions that expose demand, status quo, specificity, the narrowest wedge, observation, and future fit. Ends with a design doc written to disk.
Plan
/plan-ceo-review
CEO / founder mode. Four modes. Expansion, Selective Expansion, Hold Scope, Reduction. Finds the 10-star product hiding inside the request.
/plan-eng-review
Engineering manager mode. Locks architecture, data flow, ASCII diagrams, edge cases, and the test plan.
/ship checks the Review Readiness Dashboard for it and falls back to its own pre-landing review when it is missing./plan-design-review
Senior designer mode. Rates each design dimension 0 to 10, explains what a 10 looks like, then edits the plan to get there. Detects “AI slop” patterns.
/plan-devex-review
DX lead mode. Interactive review that explores personas, benchmarks against competitors, designs magical moments, and traces friction. Three modes. Expansion, Polish, Triage.
/design-consultation
Design partner. Builds a complete design system from scratch. Typography, color, spacing, motion. Writes
DESIGN.md as the project’s design source of truth.Build and Review
/review
Paranoid staff engineer. Finds the bugs that pass CI but blow up in production. N+1 queries, race conditions, enum gaps, escaping bugs. Auto-fixes mechanical issues. Batches judgment calls.
/investigate
Systematic debugger. Iron Law. No fixes without root-cause investigation. Three-strike rule on failed hypotheses. Auto-freezes edits to the affected module.
/design-review
Designer who codes. 80-item visual audit on a live site, then a fix loop with atomic commits and before/after screenshots.
/devex-review
DX tester. Walks the actual onboarding flow, measures time-to-hello-world, screenshots errors, compares results against
/plan-devex-review scores./design-shotgun
Design explorer. Generates AI mockup variants (default 3, up to 8 for important screens), opens a side-by-side comparison board in the browser, iterates until the user approves. Taste memory learns preferences across sessions.
/design-html
Design engineer. Generates production HTML using Pretext computed layout. Text actually reflows on resize. Heights adjust to content. 30 KB overhead, zero dependencies.
Test
/qa
QA lead. Tests a live site, fixes bugs with atomic commits, auto-generates regression tests, re-verifies. Self-regulates with a WTF-likelihood heuristic that caps at 50 fixes.
/qa-only
QA reporter. Same methodology as
/qa but report-only. Use when a pure bug list is wanted without any source-code edits./pair-agent
Cross-vendor agent orchestrator. Shares the browser with OpenClaw, Codex, Cursor, or Hermes through a scoped token plus an optional ngrok tunnel. Each agent gets its own tab.
/cso
Chief Security Officer. Fifteen-phase audit. Secrets archaeology, dependency supply chain, CI/CD security, LLM-specific risks, OWASP Top 10, STRIDE threat modeling.
Ship
/ship
Release engineer. Twenty sequential steps. Merge base, run tests and evals, audit coverage, plan-completion audit, pre-landing review, version bump, bisectable commits, push, document, open PR.
/land-and-deploy
Release engineer (post-merge). Merges the PR, waits for CI, waits for deploy, runs canary checks. One command from “approved” to “verified in production.”
/canary
Site reliability engineer. Post-deploy monitoring loop. Console errors, performance regressions, visual diffs against a pre-deploy baseline.
/benchmark
Performance engineer. Baselines page load times, Core Web Vitals, and resource sizes. Compares before and after on every PR.
Document and Reflect
/document-release
Technical writer. Cross-references the diff against every
.md file in the project. Updates README, ARCHITECTURE, CONTRIBUTING, CLAUDE.md as needed. Catches stale architecture diagrams./document-generate
Documentation author. Diataxis-framework generator. Tutorial, how-to, reference, and explanation pages produced from the code itself.
/retro
Engineering manager retrospective. Team-aware weekly summary. Commits, LOC, test ratio, hotspots, per-person praise plus growth opportunities.
/retro global runs across all projects. Stores JSON snapshots for trend lines.The eight power tools
/codex
Second opinion. Wraps OpenAI Codex CLI with three modes. Review (pass/fail gate), challenge (adversarial), consult (open conversation with session continuity). Cross-model overlap analysis when both
/review and /codex have run./careful
Safety guardrails.
PreToolUse hook that warns before rm -rf, DROP TABLE, git push --force, git reset --hard, kubectl delete, docker system prune. Override any warning./freeze
Edit lock. Blocks
Edit and Write outside a specified directory. Prevents Claude from “fixing” unrelated code while debugging./guard
Full safety.
/careful plus /freeze combined. Maximum protection for production work./unfreeze
Unlock. Removes the
/freeze boundary, returning to unrestricted edits./open-gstack-browser
GStack Browser. Launches a visible Chromium with the sidebar extension auto-loaded. Anti-bot stealth, auto model routing (Sonnet for actions, Opus for analysis), one-click cookie import.
/setup-deploy
Deploy configurator. One-time detection of the deploy platform (Fly.io, Render, Vercel, Netlify, Heroku, or GitHub Actions) and the production URL.
/gstack-upgrade
Self-updater. Detects global versus vendored install, runs the upgrade, syncs both copies if dual-installed, prints the changelog.
The fifteen supporting skills
Infrastructure and orchestration layer./browse powers every QA flow. /learn is the institutional memory every other skill writes to and queries from. /autoplan runs the full review gauntlet automatically.
/browse
Headless browser primitive. Powers
/qa, /design-review, /canary, /benchmark, /devex-review. Roughly 100 ms per call after warm-up./autoplan
Review pipeline orchestrator. Reads every plan-review skill file from disk and runs CEO, then Design, then Eng, then DX sequentially. Auto-decides via six encoded principles. Surfaces only taste decisions for user approval.
/setup-browser-cookies
Session manager. Imports cookies from Chrome, Arc, Brave, or Edge via the macOS Keychain into the headless Playwright session.
/setup-gbrain
Semantic memory onboarding. Installs gbrain (local PGLite or Supabase) and registers it as an MCP tool for cross-machine memory sync.
/sync-gbrain
Index refresh. Re-indexes the repository into gbrain’s native code surfaces.
code-def, code-refs, code-callers, code-callees./learn
Institutional memory manager. Review, search, prune, and export the per-project learnings JSONL. Other skills append to it automatically and query it via the
{{LEARNINGS_SEARCH}} template at the right moments in their prompts./plan-tune
Question sensitivity. Per-question AskUserQuestion preferences. Mark never-ask, always-ask, or only-for-one-way-doors. Observational profile of declared versus inferred behavior.
/context-save
State capture. Saves git state, decisions made, and remaining work so a future session can resume, even across Conductor workspace handoffs.
/context-restore
State restore. Loads the most recent saved context across all branches and presents the resume summary.
/health
Code-quality dashboard. Composite weighted 0 to 10 score across type checker, linter, tests, and dead-code detection. Tracks trends over time.
/landing-report
Version queue snapshot. Read-only dashboard of which
VERSION slots are claimed by open PRs and which sibling Conductor worktrees have WIP./benchmark-models
Cross-model benchmark. Runs the same prompt through Claude, GPT (via Codex CLI), and Gemini. Compares latency, tokens, cost, and optional LLM-judge quality.
/make-pdf
Publication-quality PDF. Turns any markdown file into a Faber and Faber-style PDF with 1-inch margins, page numbers, cover pages, clickable TOC.
/scrape
Read-only web scraping. First call prototypes via
$B primitives (around 30 seconds). Subsequent calls on a matching intent run the codified skill in around 200 ms./skillify
Codifier. Walks back through the conversation, locates the last
/scrape prototype, synthesizes a Playwright-via-browse-client script plus test plus fixture, and writes a permanent browser-skill on disk after the test passes.