aws-sync

Sync a local directory to or from S3, always dry-running first and running the real transfer in the background so it survives terminal/session disconnects.

Parse the argument string:

• Token 1: mode, must be upload or download. If missing or not one of these two values, ask the user which mode they want rather than guessing.
• Token 2: local_dir (a local filesystem path).
• Token 3: remote_dir (the full S3 path, e.g. s3://some-bucket/some/prefix, including the bucket). Never hardcode a bucket name or path in this skill -- the bucket always comes from whatever the user passes as remote_dir.
• Remaining tokens: optional flags. Recognize --region <value> (default to us-east-1 if not given); pass any other flags through verbatim to aws s3 sync.

Determine source and destination from mode:

• upload: source = local_dir, destination = remote_dir
• download: source = remote_dir, destination = local_dir

Derive a job name from the basename of local_dir, with any character outside [A-Za-z0-9_-] replaced by _. Use it to name log files, always under /tmp:

• dry run log: /tmp/aws_sync_<job_name>_dryrun.log
• real transfer log: /tmp/aws_sync_<job_name>.log

Steps:

1. Validate paths.
   • upload: confirm local_dir exists (ls -ld <local_dir>). If it doesn't, stop and report -- do not create it.
   • download: create local_dir if needed (mkdir -p <local_dir>) since it's the destination.
2. Dry run first, always. Run in the foreground:

   aws s3 sync <source> <destination> --region <region> --dryrun [extra flags] > <dryrun_log> 2>&1

   Then get the file count with wc -l < <dryrun_log>.
3. Report before doing anything real. Tell the user the mode, source, destination, region, and the file count from the dry run. If the count is 0, report that everything is already in sync and stop -- do not proceed to step 4.
4. Confirm before the real transfer. This moves real data into or out of cloud storage and can overwrite existing objects/files, so always confirm with the user before proceeding past the dry run -- show them the file count and both paths. Do not skip this even if the user approved a similar sync earlier in the conversation; each new local_dir/remote_dir pair needs its own confirmation.
5. Launch the real sync in the background, once confirmed:

   nohup aws s3 sync <source> <destination> --region <region> [extra flags] > <real_log> 2>&1 &
   disown

6. Give the user a progress command, using the total captured in step 2:

   watch -n 5 'echo "<job_name>: $(wc -l < <real_log> 2>/dev/null || echo 0) / <total_from_dryrun>"'

Rules:

• Never hardcode an S3 bucket name or path -- it always comes from the remote_dir argument the user supplies.
• Never skip the dry run, and never start the real transfer without explicit user confirmation shown against that dry run's numbers.
• Keep all log files under /tmp.