| name | blast-radius |
| description | Map the full blast radius of a code change before making it โ every breaking reference, transitive (ripple) caller, affected test, and historically co-changing file โ using the `br` CLI (LSP-backed, cross-language impact analysis). Use this whenever the user is about to rename, change the signature of, or remove a function, method, type, API endpoint, database column, or config/env key, or asks "what breaks if I change X", "what depends on this", "what calls this", "impact analysis", "is this safe to change/refactor", or "check the blast radius" โ even if they don't say "blast radius" explicitly. Especially reach for it before refactoring shared code. |
Blast Radius
Map everything a code change would touch before making it, so you don't ship a broken build or a silent regression. This skill drives br โ a CLI that asks the project's language server (gopls, typescript-language-server, pyright, rust-analyzer, clangd) for accurate references and call hierarchy, then layers on git history. That accuracy matters: it tells getUser in this package apart from a same-named symbol elsewhere, and it follows the call graph to find indirect callers โ neither of which a text search can do.
Prerequisite
br must be installed:
go install github.com/buddyh/blast-radius/cmd/br@latest
It also needs the language server for the code being analyzed. br doctor lists what's available and prints the install command for anything missing (e.g. go install golang.org/x/tools/gopls@latest). If br itself isn't installed, tell the user the install line above rather than silently falling back.
Run it
Identify the target type and run br analyze, always with --json so you can parse it:
br analyze "<target>" [path] --json
| Target | Example |
|---|
| symbol (function/method/type) | br analyze "getUserById" . |
| precise symbol at a location | br analyze internal/auth.go:42:6 . |
| API endpoint | br analyze "POST /api/users" . |
| database column | br analyze users.email . --kind column |
| config / env key | br analyze DATABASE_URL . |
path defaults to the current directory โ pass the repo root to analyze a different project. Kind is auto-detected (endpoints and ALL_CAPS keys); use --kind to force it. Other flags: --depth N (transitive caller depth, default 3), --lang go|typescript|python|rust|cpp, --cochange-limit N.
Read the JSON
target, def (file:line), kind, lang,
breaking[] {file, line} โ direct references
ripple[] {file, line, via} โ transitive callers (changed via the named caller)
tests[] {file, line} โ references inside test files
cochange[] {file, commits, pct} โ files that historically change together (git)
note, ambiguous[]
Present it
Give the user a decision-ready summary, not a raw dump:
- What & where โ the definition (
def), kind, and the counts.
- BREAKING โ direct references that must change with the edit. List
file:line.
- RIPPLE โ transitive callers to review for behavior changes (
file:line โ via <caller>). These compile fine but may behave differently.
- TESTS โ re-run these.
- CO-CHANGE โ files that historically move together with the target even without a static link (config read by name, sibling modules, cross-layer edits). Flag the high-percentage ones as "probably need a look too."
- Risk + change order โ small radius (โค3 breaking): change in place, update consumers in the same commit. Medium: add the new version alongside, migrate, remove. Large: strangler pattern (new symbol + deprecation, phased migration). If the target is shared across what look like service boundaries, note that other repos may be affected โ
br only sees this workspace.
If ambiguous is populated, the name matched several symbols โ tell the user and offer to re-run with file:line to pin the exact one. If note is present (e.g. a low-confidence text match for a column), surface it honestly.
When the tool can't help
If br is unavailable, or there's no language server for the language, you can fall back to a manual reference search (ripgrep the target across code, tests, docs, and config) โ but say so, because that's text-only: it can't follow transitive callers and will produce false positives on same-named symbols. Prefer installing br + the language server when the analysis actually matters.
