Files
cds-ai/cdssync/AGENTS.md
anthony.wen 7c27535e2a Add update-only mode for test dataset generator
Add support for running content updates against an existing migration
test dataset without recreating the filesystem structure.

Also make ACL/xattr updates non-fatal on filesystems that do not
support those operations.
2026-04-21 13:21:22 -04:00

62 lines
2.7 KiB
Markdown

# AI Agent Instructions for `cdssync`
These instructions apply anywhere under `/home/aw/code/cds/cdssync`.
## Migration Test Dataset Workflow
For migration test datasets in this workspace, follow this process by default:
1. Generate the dataset locally from this workspace.
2. Preserve the local generated dataset as the canonical original copy.
3. Copy the dataset to the test machine using metadata-preserving tooling.
4. Verify the copied dataset on the test machine before using it for migration testing.
## Generation Rules
- Use `/home/aw/code/cds/cdssync/generate_migration_test_dataset.sh` to create the dataset unless the user explicitly asks for a different method.
- Prefer `/home/aw/code/cds/cdssync/migration-test-dataset` as the local canonical dataset location unless the user specifies another target.
- The generator script accepts an optional `UPDATE_INTERVAL_SECONDS` argument:
- omit it to create the dataset once and exit
- use `0` for continuous random content updates
- use any integer greater than `0` to rewrite mutable files every `N` seconds
- The generator script also accepts `--update-only`:
- use it to update an existing dataset in place without recreating files, links, or directories
- combine it with `UPDATE_INTERVAL_SECONDS` to keep mutating an existing dataset on a fixed interval
- If ACL/xattr coverage matters, ensure the generation host has:
- `acl` installed for `setfacl` and `getfacl`
- `attr` installed for `setfattr` and `getfattr`
## Copy Rules
- Use `rsync -aHAX` by default when copying the dataset to another machine.
- Preserve permissions, timestamps, symlinks, hard links, ACLs, and xattrs.
- Do not use GUI copy/paste or non-preserving copy methods for this dataset unless the user explicitly asks for that.
## Verification Rules
After copying to a test machine, verify at least:
- file and directory structure
- permissions
- symlinks
- hard links
- timestamps
- ACLs
- xattrs
Preferred verification commands include:
- `find DEST_DIR | sort`
- `stat DEST_DIR/regular/script_3mb_700.sh`
- `stat DEST_DIR/readonly-dir/locked_text_1mb_444.txt`
- `readlink DEST_DIR/links/symlink_to_text_1mb_644.txt`
- `stat DEST_DIR/regular/random_3mb_644.bin DEST_DIR/links/hardlink_to_random_3mb_644.bin`
- `getfacl -p DEST_DIR/metadata/acl_text_1mb_644.txt`
- `getfattr -d DEST_DIR/metadata/xattr_text_1mb_644.txt`
## Destination Host Requirements
- If the destination host lacks `acl` or `attr`, ACL/xattr verification will be incomplete.
- If the destination filesystem does not support ACLs or xattrs, those attributes may not survive transfer even when the copy method is correct.
- The generator now logs and continues when ACL/xattr assignment is unsupported on the target filesystem instead of exiting.