Prompt Deck

Create a Claude Code slash command /pull that takes a repo name, finds that repo under my dev roots (~/Claude and ~/workspace, including nested folders and worktrees), pulls from remote, merges, and auto-resolves trivial conflicts while preserving my uncommitted work. Model it on my existing /sync skill. Show me the command file before installing it.

Set up a launchd job on this Mac that runs my /sync across every iserlabs repo under ~/Claude and ~/workspace every morning at 6am, logs the result, and notifies me only if a repo had a merge conflict it could not auto-resolve. Walk me through installing and loading it.

Add a Claude Code hook that, when a session passes 40 tool calls or before a /clear, automatically writes a resume snapshot the way my /save skill does, so I never lose context on a long session. Show me the settings.json change first.

Before you start this task, interrogate me with clickable multiple-choice questions until you have everything you need. Then execute the entire task end to end without stopping to ask 'shall I continue'. Only stop for a real decision fork, a blocker, or an irreversible action. The task is: <describe your task>.

Look at my last 10 sessions and count how many of my messages were just 'go', 'continue', 'keep going', 'do it', or 'yes'. Tell me which tasks generated the most nudges and rewrite the opening prompt I should have sent so the whole thing would have run in one shot.

This is a quick, mechanical task, not deep reasoning. Use Haiku for it to save cost, and tell me if partway through you think it actually needs Opus. Task: <describe>.

Read my usage-lens metrics.json and list the 10 sessions where I spent the most Opus-equivalent value on tasks that a smaller model could have handled. For each, say what signal (few messages, simple edits) makes it a candidate, and what I should have done instead.

I have several related tasks in the same codebase. Do them all in this one session so the context is cached once instead of re-primed per session. Here is the list: <list the tasks>.

Here are the Claude Code skills I have installed but never used: <paste the unused list from the dashboard>. Pick the 5 that would most help my Next 16 Ops Hub and my STR marketing sites, and for each show me a concrete example prompt that uses it on my actual work.

I frequently fan out subagents for audits and migrations by hand. Take my most common multi-agent task and turn it into a Vercel-style workflow with structured fan-out, retries, and resume, so I stop herding agents manually. Start by asking me which task to convert.

It is late and my error rate climbs at night. For this task, flag every risky or irreversible step (migrations, prod deploys, domain or DNS changes, git history) and hold those for me to run in daylight. Do the safe, reversible parts now. Task: <describe>.

Read my ~/.claude/CLAUDE.md and suggest 3 specific edits that would reduce how often I have to say 'go' or 'continue', based on the tasks where I nudge the most. Show the exact before and after lines.