Up To Date

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 ") and proceed. Proactive runs stay strictly read-only — the sync decision (step 7) still requires explicit OK.

Steps

1. Confirm it's a git repo with a remote. If not, say so and stop — there's nothing to sync.
2. Local state: git status --short and git stash list. Note uncommitted changes, untracked files, stashes. This gates everything below.
3. Fetch: git fetch --all --prune. (Fetch is safe; it changes nothing local.)
4. Divergence: current branch vs its upstream — git rev-list --left-right --count @{upstream}...HEAD. Report ahead/behind. No upstream (new local branch, or detached HEAD) → that command fails; say so plainly, fall back to origin/<default-branch> for the comparison, and note the branch is unpushed. Never report "in sync" from a failed command.
5. Recent activity: git log --oneline -15 on the branch, and the same for the default branch if you're not on it. Summarize what changed, not every line.
6. Open work (if gh is available and the remote is GitHub): gh pr list and gh issue list. Surface anything touching the files you're about to work on.
7. 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.
8. Library currency (only if asked or docs arg 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 gh is 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 status answer 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 → architect next, 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.