The Shipping Set

After this, the AI stops jumping from "sure, let's build it" to "here are 800 lines of code" — and starts running the actual workflow: PRD, plan, tickets, isolated execution, proof of completion.

Who this is for

Engineers shipping non-trivial features with an AI assistant. You've been burned by "finished" work that broke production, plans that skipped the awkward question, agent runs that started before the design was settled. You want the assistant to behave like the senior engineer who insists on the gates.

Before — default Claude, prompt: "Add a feature flag system"

Streams 600 lines of code: flag service, admin UI, SDK, four storage backends. Half of it built against assumptions (per-user? per-org? gradual rollout? kill switch?). Three hours later, you've ripped out the SDK and started over.

After — same prompt, with Shipping Set installed

Step 1 — PRD from context: 8 questions (granularity, kill-switch latency, auth integration, etc.) → one-page PRD.
Step 2 — Implementation plan: architecture options A/B/C with trade-offs, picks one against your constraints.
Step 3 — Slice into tickets: 8 tickets, ~4 hours each, dependencies marked.
Step 4 — Isolate the work: new branch, empty scaffold, ready for tickets.
Step 5 — Run the plan: picks ticket 1, implements, runs verification, stops at any blocker.
Step 6 — Claim with proof: "Ticket 1 done. Tests: 12 pass, 0 fail. Coverage: 89%. Diff: 240 lines."

What's inside

• prd-from-context — turns "add X" into a one-page PRD with the awkward questions answered
• implementation-plan — designs the architecture with options + trade-offs before any code
• slice-into-tickets — breaks the plan into 4-hour tickets with dependencies marked
• isolate-the-work — sets up the branch, scaffolds, prevents work bleeding across features
• run-the-plan — executes tickets one at a time, stops at the first blocker, doesn't skip ahead
• claim-with-proof — runs verification before saying "done," shows the receipts inline

Three ways to use this

1. Paste it (any AI, 30 seconds, no install)
   Open the .md file, copy contents, paste into Claude / ChatGPT / Cursor / Perplexity with: "Read this and apply it whenever the task it describes comes up in our conversation." Done.
2. Install it (Claude Code / Cursor users)
   Drop the unzipped folder into ~/.claude/skills/ (Claude Code), .cursor/skills/ (Cursor), or your IDE's equivalent. Auto-loads next session.
3. New to AI assistants? Get Claude free at claude.ai, then use Option 1.

€24 — six skills · sum of parts €42