Sync Project

Sync Project

Arrival-side sync for projects worked on across multiple machines (typically local macOS and Yale HPC cluster). Ensures git, conda, and renv are current before starting work.

This skill is arrival-only. Departure-side tasks (env export, commit, push) are handled by /done.

Step 0: Detect Environment

Determine which machine we're on:

# Check for module command (cluster indicator)
type module &>/dev/null && echo "cluster" || echo "local"
hostname

Set the conda activation pattern based on environment:

• Local (macOS): source ~/miniconda3/etc/profile.d/conda.sh
• Cluster (Bouchet): module load miniconda && source $(conda info --base)/etc/profile.d/conda.sh

Read the project's .claude/CLAUDE.md to find:

• The conda environment name (from the Environment section)
• Whether renv is used

Step 1: Git Sync

1a. Check local state

git status --short

If uncommitted changes exist, stop and warn:

"You have uncommitted changes. Commit or stash before syncing?"

Do NOT proceed with git pull if there are uncommitted changes — this risks merge conflicts that could lose work.

1b. Fetch and classify divergence

git fetch
git rev-list --left-right --count @{u}...HEAD   # → "<behind> <ahead>"

Branch on the counts (do NOT blindly git pull — it silently merges on divergence):

Report what was updated (files changed, new files, deletions).

1c. Diverged — choose integration (rebase vs merge)

Reached only when both sides have commits.

1. Preview conflicts (non-destructive — changes nothing):

   git merge-tree --write-tree @{u} HEAD   # exit 0 = clean; exit 1 = conflicts (lists files)
   

2. Recommend an approach:
   • Local commits not yet pushed → rebase (git pull --rebase): replays your commits on top of the remote → linear history. Safe because unshared commits can be rewritten. ← default recommendation.
   • Local commits already pushed / shared → merge (git pull): creates a merge commit; never rewrites shared history.
3. Report the preview result + recommendation, then proceed only on user confirmation.
4. If the chosen integration hits conflicts, report them and stop — do NOT auto-resolve.

Step 2: Conda Environment

2a. Check if environment.yml exists

If no environment.yml in the project root, skip this step.

2b. Read env name from environment.yml

head -1 environment.yml  # name: my-env

2c. Check if environment exists

conda env list | grep -q "env-name"

• Env doesn't exist: Create it: conda env create -f environment.yml
• Env exists: Update it: conda env update -f environment.yml --prune

The --prune flag removes packages no longer in environment.yml.

2d. Verify

conda activate env-name && python --version

Report the Python version and package count as a sanity check.

Step 3: renv (if applicable)

Skip if no renv.lock in the project root.

3a. Check status

Rscript -e "renv::status()"

3b. Restore if needed

If packages are out of sync:

Rscript -e "renv::restore()"

Report what was installed/updated or "renv is in sync."

Step 4: Memory Promotion

Scan for project-relevant memory files:

ls ~/.claude/projects/*<project-name>*/memory/ 2>/dev/null

Where <project-name> is a fuzzy match on the project directory name.

If memory files exist:

1. Read each one
2. For each, assess: does this contain information that should be in a .claude/ planning doc or CLAUDE.md so it's available on both machines?
3. If yes, show the memory content and suggest where to add it
4. Ask the user before making changes

If no memory directory exists or it's empty, skip silently.

Step 5: Report

/sync-project complete

Git:    pulled 3 files (scripts/01_busco.qmd, batch/run_busco.sh, ...)
Conda:  spongilla-genome updated (2 packages added)
renv:   in sync
Memory: 1 memory promoted to .claude/STRESS_TEST_FINDINGS.md

Ready to work.

environment.yml Hygiene

These checks run as part of Step 2 but are documented here for reference.

When this skill reads environment.yml, flag these issues:

These are warnings only — the skill does not auto-fix environment.yml during arrival. Fixes happen at departure time (in /done) when exporting the environment.

What This Skill Does NOT Do

• Does not commit, push, or export environments (that's /done)
• Does not handle ~/.claude sync (that's /sync-cluster)
• Does not check batch script provenance blocks (arrival is not the time for that)
• Does not auto-run — must be explicitly invoked with /sync-project