marekpolcar
- 4 skills
- 0 followers
- 11 hours ago last updated
- ▌ Solo · marekpolcar bundleUnattended run from a spec - or from instructions typed straight into the invocation, or already settled in this conversation, in which case the spec gets written for you. Plan, execute, decide, log every call for audit, and survive a session-limit hit: the run checkpoints to disk and a launchd watcher boots a fresh session from it. Closes with an HTML run report - timeline, decisions, hiccups - and optionally a delivery check and a correctness review that fix what they find. /solo <spec-path|instructions> emits a mandate to paste into a fresh session; /solo go ... runs it here, now.
- ▌ Handoff · marekpolcarUse at the end of a session when you want a single self-contained prompt that bootstraps a fresh, zero-context conversation - switching chats, clearing context, hitting the context limit, or moving to another machine, with no runbook or audit trail in play. Produces a prompt a brand-new Claude (no memory of this session) can act on immediately. Triggers: "handoff", "/handoff", "write the prompt for a new chat", "I'm clearing context - give me a prompt to continue", "hand this off to a fresh session". Not for handoffs partway through a procedure that keeps its own on-disk audit trail - use whatever skill links that trail instead.
- ▌ Check Plan · marekpolcar bundleReview and validate an implementation plan or a spec/design doc before it is built, using a mandatory repository ground-truth pass and a panel of parallel reviewers. Use whenever a plan is about to be executed, whenever a spec is about to be turned into a plan or into code, when either was written more than a day ago, or when the user invokes /check-plan. Applies to any domain - implementation, migration, refactor, research, design, ops.
- ▌ Check Delivery · marekpolcar bundleReview a finished implementation against the plan or spec it was built from. Runs the project's own verification gates first, then a panel of parallel reviewers over the diff. Use when an implementation is complete, before merging or releasing, when a change needs a second opinion that is not just a reading, or when the user invokes /check-delivery. Every behavioural finding must cite a command, a test, or a call site.