v1.1.0

The review loop gets a skill of its own, and the three agent skills drive their CLIs in any role.

New: independent-review (in ai-shipping). The unsteered review before anything ships: a fresh reviewer that never saw the build reads the unreleased diff and the working tree, round after round, until a round finds nothing the change introduced above Low. It used to live in grok-agents, as if Grok were the review. Grok, then a fresh read-only Claude Code session, are now stated as the defaults they are, and your agent instructions can name another reviewer, Codex included. Its wrapper is scripts/review.py (it was grok_review.py).

Changed: the agent skills, one shape for all three

• grok-agents: the Grok CLI driven from another agent as a peer, a builder or a fresh reviewer, with sessions you name and resume, the isolation a review needs, and sign-in by device code.
• codex-agents: gains a reviewer mode, a read-only Codex session that leaves your Codex config and its tools out.
• claude-agents: unchanged in role; it points reviews of commits to independent-review.

If you named grok-agents in your own agent instructions for the review loop, point that line at independent-review. Nothing was renamed, so the old name still resolves, now to the Grok driver, which sends reviews to the new skill.

README: it now says what you decide and review (the proposal, the plan, one page of what waits on you, the release's evidence), not only what you click, and it's shorter. With more than one agent, only the coordinating one needs the skills: a builder works from its brief, and a reviewer starts without them.

44 skills in six packs. The README, the walkthrough and the process graphic list the new skill.

Install: README § Install. Lost at any point? Type /product:start-here (Codex: $product:start-here) and say what you have.

v1.0.0

Agent skills for shipping real products with coding agents, from the first idea to paying users. No coding.

The first tagged release: 43 skills in six packs (product, ai-shipping, agent-teams, agent-harness, app-stack, work-humanizer). You start four (start-here, ship-check, handover, retro); the other 39 start themselves when the work matches.

New since the collection went public

• legal-pages: terms, a privacy policy and consent drafted from what the code collects and sends, with the facts only you have asked for, and a test that holds them to the code.
• product-analytics: session recordings and heatmaps without analytics cookies (Microsoft Clarity by default, PostHog optional), read back by your agent over MCP.
• One process, the same in every skill: build with its reviews, your click-through, a release review where the deeper checks are proposed and you pick, then the release you say go to.
• Install in one prompt, in Claude Code or the Claude desktop app, with the commands for Codex and other agents beside it.
• The packs are unpinned, so an install with auto-update follows main.
• A Tests workflow runs the collection's own checks on every push: every skill and manifest, the always-loaded descriptions' budget, and claude plugin validate.

Learned from a second product built with the skills

• Uptime monitoring proposes Staygreen by default (made by the author of this collection); a monitoring service you already use stays.
• Supabase: the CLI is never linked. Every database, development's included, is reached by its own connection string, so your agent needs an access token only for the optional, read-only MCP. Supabase CLI 3.0.0 (out today) renames the flags; the skill says to pin the CLI.
• A fresh Next.js 16.4 app turns Cache Components on, which the stack's rules assume off; the build stage now scaffolds with it off.
• Review loops: when one mechanism draws findings two rounds running, the skill says to stop patching it: cut it, redesign it once, or pause it behind a switch that is off by default. When Grok is out of credit, the review wrapper says so and prints the same round for a fresh Claude session.
• Tests that catch what reviews missed: every page rendered once for each role, every new database function run end to end, a fully populated workspace deleted.
• Calls to URLs your customers give you (server-side request forgery) and their secrets stored at rest, authorship in a multi-tenant app, Clerk setup through its Backend API, and deploys made by CLI upload.
• The skill scripts run on macOS's stock Python 3.9.

Independent review rounds continue on every change; fixes ship as patch releases.

Install: README § Install. Lost at any point? Type /product:start-here (Codex: $product:start-here) and say what you have.