rescue-stranglerlisted
Install: claude install-skill vivekbhange-hub/legacy-rescue-skill
# rescue-strangler
Rewrites from scratch fail in a familiar way: two years in, the new system still lacks the
edge cases the old one learned the hard way, and the old one kept changing underneath. The
alternative is to grow the new thing around the old one, move traffic over a slice at a time,
and delete the old code only when nothing calls it. This skill produces that plan for one
component, grounded in the actual call sites, with every milestone shippable and reversible.
## Inputs
- `$ARGUMENTS`: the component to replace (file, module, endpoint group, table, library) and
the target if known.
- `rescue/map.md`, `rescue/hotspots.md`, `rescue/characterization.md` if present.
- Constraints from the user: deadline, freeze windows, team size, whether two versions can
run side by side, data ownership.
## Output
- `rescue/strangler-<component>.md`: the migration plan.
- No code changes. Implementation happens milestone by milestone through `rescue-refactor`
and normal feature work.
## Steps
1. **Draw the current boundary.** Find every caller of the component: Grep for imports,
class or function names, endpoint paths, table names, queue names. Group callers by entry
point. Count them; this is the surface the migration has to cover. Note callers outside the
repo (other services, mobile apps, cron, reports) from config and docs, and ask the user
about ones you cannot see.
2. **Choose the pattern** with these rules:
- **Strangler fig (routing)**: the compon