seansmithworks
UserSkills I use to run Claude Code as a designer — wrap family + gh cleanup family
Categories
Indexed Skills (10)
gh-clean-branches
Prune stale local branches — detects branches whose remote is gone and merged feature branches still hanging around. Shows a combined list for confirmation before deleting anything.
gh-fork-sync
Sync your fork's main branch with upstream. Shows how far behind you are, confirms the strategy, syncs and pushes to your origin. Safety guard — never pushes to upstream.
gh-pr-triage
Survey open PRs — yours, assigned, and review-requested — grouped by what action is needed. Outputs a table with a concrete next-action suggestion per PR.
gh-stale-issues
Triage open issues by age and last activity. Groups into action buckets and suggests what to do per issue. Read-only — surfaces actions, doesn't take them.
orchestrator-route
Route an over-cap ORCHESTRATOR.md down to a router-shaped file without destroying durable value. Use when a size gate fires, when /wrap reports the file is over its 10,000-15,000-char target range (20,000 hard max), or when Sean says "route this", "prune orchestrator", "this file is too big", "/orchestrator-route". NOT for a file that is merely untidy — the trigger is size or staleness, and the operation is routing facts to their owners, never deleting by age.
orchestrator-scaffold
Scaffold missing project files via three modes. Use when the user says "/orchestrator:scaffold", "scaffold this project", "set up agent files", "create grounding files", "add mission brand principles", or when Phase 1 boot check surfaces missing scaffolding. Three modes — default (auto-detect), lifecycle (v2.1 agent files), grounding (v2.2 mission/brand/principles interview).
orchestrator-update
Review and apply pending orchestrator config updates. Use when the user says "/orchestrator:update", "check pending updates", or on session start if the registry has pending entries. Surfaces improvements from `~/.claude/PENDING-UPDATES.md`, applies selected ones via subagents, and validates with the eval suite.
wrap-continue
Checkpoint a long-lived thread mid-task so it can be thrown away and resumed in a fresh one — stop background writers, commit, smart-push, reconcile open items into BACKLOG.md, and render a copyable pickup prompt inline. Use when the work is NOT finished and continues immediately in a new thread ("wrap and continue", "checkpoint", "context reset", "trim and keep going", "recycle this thread"). Do NOT use when the session is actually over — done for the day, work complete, or switching to unrelated work; that is /wrap's heavy retrospective. The distinguishing test is whether the task is finished, not whether the thread is ending.
wrap
End-of-session close-out — secure uncommitted work, put every open item in a durable home, refresh shared project state, and report honestly what was captured and what was skipped. Use when Sean is done for the day or longer ("wrap", "wrap it up", "done for now", "that's it for today", "end session", /wrap), including when the session contained almost nothing — it then correctly does almost nothing rather than not running. Retrospective only — never emits a resume/pickup prompt. Do NOT use for mid-task context recycling where the same work continues in a fresh thread (that is /wrap-continue), for answering questions about what wrapping does, or when a single narrower capture was asked for (just commit this, save one note).
orchestrator-boot
Run the orchestrator session-start check — task file, pending updates, project scaffolding, context reads, rehydration. Use on first activation of an orchestrator thread, or when Sean says "boot", "orchestrator boot", "start up", or "get up to speed on this project".
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.