cuga-project CUGA Source Sync
Scout a source repo and write (or update) a CLAUDE.md at its root so the cuga-migrator
skill's analyst stage can read it first and go straight to relevant files without wasting context
on large test fixtures or build artifacts. Run this once per source repo, before cuga-migrator.
Arguments
Parse the user's request for <source-name>.
Source repo path: migration_from/<source_name>/
Output: migration_from/<source_name>/CLAUDE.md
List the top-3-level directory tree under migration_from/<source_name>/, excluding .git,
node_modules, and __pycache__. Identify the top-level subfolders and their roles.
List files by line count, largest first. Flag any file over 300 lines living under a noisy directory pattern as a skip candidate:
| Pattern | Reason to skip |
|---|---|
tests/, test/, __tests__/ |
Unit/integration tests — not agent logic |
*/mocks/, */fixtures/, */testmode-data/ |
Pre-recorded test fixtures |
node_modules/, dist/, build/, .next/ |
Vendored dependencies / build output |
licenses/, LICENSE*, non_ibm_license* |
License text — not needed |
*.lock, uv.lock, pnpm-lock.yaml |
Dependency lock files |
*.db, *.pyc, *.pyo |
Binary / compiled artifacts |
*.html, *.png, *.pdf |
Non-source assets |
If README.md or README.rst exists at the source repo root and is under 300 lines, read it.
Extract: what the system does, its main subfolders/repos and their roles, the runtime flow (how
components talk to each other), how to run it.
If a CLAUDE.md already exists at the source root, read it too — you will update it, not
overwrite blindly.
Scan for: main.py, server.py, app.py, entrypoint.py; *agent*.py, *supervisor*.py,
*graph*.py; *agent*.yaml, *config*.yaml; .env.template, .env.example;
requirements.txt, pyproject.toml, package.json; one level of src/ contents. List these as
"key files to read" in priority order.
Write migration_from/<source_name>/CLAUDE.md with this structure. If it already exists, update
it — preserve manually written sections that are still accurate, replace stale ones. Write real
content throughout, no placeholders:
# <source-name> — Source System Overview
<2-4 sentence description of what the system does>
## Subfolders / repos
### `<subfolder-1>` — <role>
- **What it is:** <one line>
- **Framework:** <language + framework>
- **Key files:** <entry points>
- **Config:** <config file>
- **Run:** <how to start it>
## Relationship
<How components talk to each other — what calls what, the data flow>
## Files to skip — DO NOT READ
| Path | Reason |
|------|--------|
| `<path>` | <why> |
## Key files to read (in order)
1. `<path>` — <why>
## Migration target
<One sentence on what maps to what in CUGA>
Print a one-paragraph summary: what directories were found, how many files were flagged to skip, which entry points were identified, and whether the CLAUDE.md was created or updated.