Up to date
Six Claude Code skills (architect, review-swarm, ask-the-council, prompt-generator, up-to-date, ponytail) bundled as an installable plugin.
npx -y skills add alonbaron/claude-skills --skill up-to-dateAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 6 stars6 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Preflight sync + situational brief before repo work: fetch origin, ahead/behind divergence, recent commits, open PRs/issues, dirty-state warnings, one recommended first action. Read-only — never pulls or rewrites a dirty tree without explicit OK. Use proactively before starting work in any repo with a remote, especially shared or long-untouched ones. Also on "up-to-date", "sync first", "pull latest", "catch me up on the repo". Not for repos without a remote or quick mid-task re-checks.
SKILL.md
4.2 KB, as published. Nobody here has run it
Up To Date
Get current before doing work. Goal: a short situational brief so you build on the latest code with full context. Read-only by default. Never mutate a dirty tree, never force, never discard local work without explicit confirmation.
Proactive use
When work is about to start in a repo with a remote, invoke this without being asked: announce in one line ("Running up-to-date on <repo>") and proceed. Proactive runs stay strictly read-only — the sync decision (step 7) still requires explicit OK.
Steps
- Confirm it's a git repo with a remote. If not, say so and stop — there's nothing to sync.
- Local state:
git status --shortandgit stash list. Note uncommitted changes, untracked files, stashes. This gates everything below. - Fetch:
git fetch --all --prune. (Fetch is safe; it changes nothing local.) - Divergence: current branch vs its upstream —
git rev-list --left-right --count @{upstream}...HEAD. Report ahead/behind. - Recent activity:
git log --oneline -15on the branch, and the same for the default branch if you're not on it. Summarize what changed, not every line. - Open work (if
ghis available and the remote is GitHub):gh pr listandgh issue list. Surface anything touching the files you're about to work on. - Sync decision:
- Clean + behind only → offer
git pull --ff-only. Do it only on confirm. - Diverged (ahead and behind) → explain; recommend rebase or merge; ask.
- Dirty → do NOT pull. Report the dirty state and let the user decide.
- Clean + behind only → offer
- Library currency (only if asked or
docsarg given): for the key deps in play, use context7 (resolve-library-id → query-docs) to pull current API docs, so code targets today's API, not stale memory.
Output: the brief
- Branch + sync state (ahead/behind, clean/dirty) in one line.
- Recent commits — 2–4 line summary of what's new.
- Open PRs/issues relevant to the intended work (or "none touching this").
- Warnings — dirty tree, diverged history, stale deps.
- Recommended first action — one line.
Rules
- Report only what commands return. Never invent commit messages, PR titles, or authors.
- No mutating command (
pull,rebase,merge,checkout,stash pop) without explicit user OK. - If
ghis missing or unauthenticated, skip that section and say so — don't guess at remote state.
When not to use
- No remote → nothing to sync; say so in one line and start the work.
- Mid-task re-checks in a repo already briefed this session —
git fetch+git statusanswer it without the full brief.
Hand-offs
- Brief delivered → start the actual task on the now-current tree.
- The task is a new app/feature/workstream with no design docs →
architectnext, before code.
Done when
The user knows: what branch they're on, whether it's behind/ahead/clean, what changed recently, what open work overlaps, and the one safe next action.
Global rules
Apply on every run — canonical home ~/.claude/CLAUDE.md:
- Ground everything. Only what's given or verified; never invent files, APIs, or facts. Unknown → say "I don't know" or state the assumption.
- No tokenmaxing. Lead with the answer; keep an output budget; no filler, no restating the question.
- Agent discipline. Read before you edit; small reversible changes; ask when blocked, don't guess; report failures honestly.
- Commits are the user's alone. Author = the user; never add an AI co-author or
Co-Authored-By/credit line, and don't mention AI in commit messages.