From ship-with-ai
Use this skill whenever the user wants to ship a side project end-to-end on a deterministic, repeatable agent harness. Also trigger when the user says "/autopilot", "ship my side project on autopilot", "set up the harness", "wire up Archon for me", or "run the 5-decision setup". Takes idea + target user + stack as input, scaffolds three input files (idea.md/user.md/stack.md) plus done.md per feature, installs Archon, drops in the canonical SCOPE→BUILD→DEPLOY→FIRST-USER workflow with three approval gates, and runs it end-to-end. Human-in-the-loop: agent pauses at each gate (continue/revise/kill). Never auto-pushes to live URL without DEPLOY-gate approval. Appends one new rule to the project CLAUDE.md after every run.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ship-with-ai:autopilotThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
The reusable 5-decision harness that turns every side-project session from improvisation into execution. One Saturday to wire. Every project after runs the same way.
The reusable 5-decision harness that turns every side-project session from improvisation into execution. One Saturday to wire. Every project after runs the same way.
Ship With AI artifact #24 · post-LtM methodology · companion to How To Build The Harness That Ships For You.
Invoke when the user:
/autopilot or /autopilot "<idea in one sentence>"Do NOT invoke when the user wants to:
/autopilot → operate on cwd (must be a git repo with at least one commit, or a fresh empty directory)/autopilot "the idea in one sentence" → operate on cwd with the idea pre-filledIf cwd is not a git repo and no path is given, run git init first and warn the user.
Read, in order, and form an internal picture:
git status and git log --oneline | head -5 — confirm it's a git repols -F — see what's already in the directoryidea.md / user.md / stack.md / done.md — note overwrite risk.archon/workflows/ — note whether Archon is already wired hereCLAUDE.md for existing ## Lessons section — preserve and append, don't overwriteDo not print anything user-facing during this phase.
Ask these three questions, in this order, and wait for the user to answer each before proceeding. Do not ask follow-up questions. Three questions is the budget.
Q1. What are you building? (One to three sentences. Plain English. No tech detail yet.)
Example: "A tool that lets a parent generate a custom bedtime story for their kid in under 60 seconds, using the kid's name and 3 favorite themes."
Q2. Who is the one specific first user? (A real person if possible. Job, context, what they currently use instead, what would make them switch.)
Example: "Sara, my sister-in-law, mom of a 4-year-old, currently reads from a stack of 6 worn books, would switch if she could give her daughter a story with her own name in it."
Q3. What's your stack default? (Language, framework, deploy target. One line each.)
Example: "TypeScript, Next.js 14 app router, Vercel."
All other calibration comes from Phase 1 discovery. Do not ask follow-ups.
idea.md from Q1 (3 sentences max — see template below).user.md from Q2 (1 named user, structure who/currently/switch-trigger).stack.md from Q3 (3 lines: language/framework/deploy).archon is not on PATH, run bash ${SKILL_DIR}/scripts/install-archon.sh and wait for it to finish.${SKILL_DIR}/assets/ship-side-project.yaml into .archon/workflows/.go before proceeding to Phase 4.If Archon is already installed at the system level, skip step 4 and print "Archon already installed at $(which archon)."
On the user's exact reply of go:
/workflow approve <run-id> or reject with feedback at each."archon workflow run ship-side-project "<one-line task description from idea.md>" — Archon takes over from here.approval: node pauses with a gate_message and waits for /workflow approve <run-id> (or reject). The skill is done once Archon hands control back at the patch step.Do not auto-confirm any gate on the user's behalf. Even if the agent thinks the artifact is fine.
Do not modify the workflow YAML mid-run. The harness is the constant.
After Archon returns:
CLAUDE.md under a ## Lessons section (create if missing).# Idea
{One sentence: what it does.}
{One sentence: why it's worth building.}
{One sentence: what makes it different from what already exists.}
# Target User
**Who:** {name + role/context, e.g. "Sara, mom of a 4-year-old"}
**Currently does:** {what they use instead today}
**What would make them switch:** {the one specific trigger}
# Stack Defaults
**Language:** {e.g. TypeScript}
**Framework:** {e.g. Next.js 14 app router}
**Deploy:** {e.g. Vercel}
# Done — {feature name}
- {acceptance line 1, machine-checkable wherever possible}
- {acceptance line 2}
- {acceptance line 3}
## Lessons (added {today as YYYY-MM-DD})
- {one rule extracted from this run's failure modes — phrased as a positive
directive the agent can act on, not as a negative complaint}
vercel deploy --prod (or any deploy command) without DEPLOY-gate confirmation. Live URLs are visible. Confirm first.## Lessons section. Always append.The skill completed successfully when:
idea.md + user.md + stack.md exist in the project root.archon/workflows/ship-side-project.yaml is in placeCLAUDE.md ## Lessons sectionThe user should now be able to:
archon workflow run ship-side-project "<task>" against any future side project with a fresh idea.md + user.md + stack.mdBegin with Phase 1 (silent discovery). Output nothing user-facing until Phase 2 (the three questions).
npx claudepluginhub ship-with-ai/skills --plugin ship-with-aiProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.