Claude-starter cleanup-defensive

Remove try/catch and defensive null-checks that don't serve a real role. The goal is errors that propagate cleanly to the boundary that knows how to handle them, not silent fallbacks that hide bugs.

Core principle: catch only when you can do something more useful than the default propagation — and "log and rethrow" is rarely more useful than just letting it throw.

Preflight

1. Language detect: TS/JS, Python, Go (

   if err != nil

   patterns), Rust (

   unwrap_or

   /

   map_err

   chains that smell defensive).
2. Git state: refuse on dirty tree.
3. Report dir: ensure exists.
4. Identify boundary files — preserve catches in:
   • HTTP request handlers (

     app/api/

     ,

     routes/

     , Hono/Express/FastAPI handlers)
   • CLI entry points (files with

     if __name__ == "__main__"

     ,

     bin/*

     ,

     cmd/*

     )
   • Message/queue consumers (worker entry points, cron handlers)
   • Test files (test runners need failures localized)

Detect

TypeScript / JavaScript

# Find every try/catch
grep -rn --include="*.ts" --include="*.tsx" --include="*.js" -B1 -A5 "try {" \
  --exclude-dir=node_modules --exclude-dir=.next . > /tmp/try-catches.txt

# ESLint can flag the obvious useless ones
npx eslint --rule '{"no-useless-catch": "error"}' --no-eslintrc . 2>&1 | grep no-useless-catch > /tmp/useless-catch.txt

Categorize each catch block by content:

• Rethrow only:

  catch (e) { throw e }

  or

  catch (e) { throw new Error(...) }

  with no context added
• Swallow + null/empty return:

  catch { return null }

  ,

  catch { return [] }

  ,

  catch { return {} }

• Swallow + log only:

  catch (e) { console.error(e) }

  then continue silently
• Log + rethrow:

  catch (e) { logger.error(e); throw e }

  — usually pointless if a global handler logs
• Substantive: actually does cleanup, retries, returns a typed Result, transforms the error meaningfully

Python

grep -rn --include="*.py" -B1 -A5 "try:" . > /tmp/py-try.txt
# Bare except is always defensive
grep -rn --include="*.py" -E "except\s*:|except\s+Exception\s*:" . > /tmp/py-bare-except.txt

Go

# Find error-swallowing patterns: ignored errors, generic fallbacks
grep -rn --include="*.go" -E "_ = .*\.Err|_, _ =" . > /tmp/go-ignored.txt

Rust

grep -rn --include="*.rs" -E "\.unwrap_or\(|\.unwrap_or_default\(|\.ok\(\)" . > /tmp/rust-defensive.txt

Assess

Write

.claude/cleanup-reports/cleanup-defensive-{YYYY-MM-DD}.md

:

# Defensive Code Assessment — YYYY-MM-DD

## Summary
- try/catch blocks scanned: N
- HIGH (remove): X — pure rethrow, swallow-and-return-null, useless wrappers
- MEDIUM (review): Y — log+rethrow, broad excepts with context
- LOW (preserve): Z — substantive handling, boundary code

## Findings

### HIGH — `lib/parse.ts:45`
```ts
try {
  return JSON.parse(s)
} catch (e) {
  return null
}

Problem: caller can't distinguish "valid JSON null" from "parse failed". Hides bugs. Fix: remove try; let JSON.parse throw. If callers need optional, change return to

Result<T, ParseError>

or have caller wrap.

HIGH —

services/user.ts:88

try {
  return await fetchUser(id)
} catch (e) {
  throw e
}

Problem: literally a no-op wrapper. Fix: remove try/catch entirely.

MEDIUM —

services/payment.ts:120

try {
  await charge(amount)
} catch (e) {
  logger.error('charge failed', { e, userId })
  throw e
}

Problem: log+rethrow. If global handler also logs, this is duplication. Recommendation: check if there's a global error logger. If yes, remove. If the contextual data (

userId

) isn't otherwise captured, keep but add a note.

LOW —

app/api/[...path]/route.ts:34

— preserve, this is the API boundary.

Critical Assessment

[2-3 paragraphs on patterns: is this codebase prone to silent fallbacks? Are there layers wrapping errors unnecessarily?]

## Apply

**Auto-remove HIGH only.**

### Confidence rubric

**HIGH (auto-remove):**
- `try { x } catch { } ` (silent swallow, no return)
- `try { x } catch (e) { throw e }` (no-op)
- `try { x } catch (e) { throw new Error(e.message) }` (loses stack, adds nothing)
- `try { x } catch { return null/undefined/[]/{}/0 }` (silent fallback that hides errors) — only if file is NOT a boundary file
- Python `except: pass` blocks
- Go: `_ = someCall()` where the discarded value is `error`

**MEDIUM (report only):**
- Log + rethrow (might be intentional observability)
- Catch with retry logic
- Catch in middleware (might be intentional last-resort)
- `unwrap_or(default)` where default is a real value (might be intentional)

**LOW (preserve):**
- Inside boundary files (see preflight list)
- Inside `__init__.py` import-time error handlers (compatibility fallbacks)
- Catches that transform exception type meaningfully (e.g., `catch DBError { throw new ValidationError() }`)
- Cleanup catches with `finally` doing real work

### Execution

For each HIGH:
1. Remove the try wrapper, keep the body's expression.
2. If the function signature implied "may return null on error" because of the catch, that signature is now lying — flag for human review (do NOT change signatures automatically).
3. Single commit: `chore(cleanup): cleanup-defensive — removed N useless try/catch blocks`.

## Verify

```bash
npm run check 2>&1 || (npx tsc --noEmit && npx eslint .)
npm test 2>&1
pytest 2>&1
go test ./... 2>&1
cargo test 2>&1

Tests are critical here — removing a swallowed error often surfaces a real bug that was being hidden. If tests fail:

1. Read the failure carefully — is it a real bug being surfaced (good, don't revert; flag for human), or did we remove a catch that was load-bearing for tests (revert that one).
2. Default action: revert and downgrade to MEDIUM. The human can decide whether the surfaced failure is "actually a bug we should fix."

Output

• "Removed N useless try/catch blocks. M deferred for review."
• Report path.
• Verify status. If tests surfaced previously-hidden errors: highlight clearly with "⚠️ Real errors surfaced — see report."

NEVER

• Remove a catch in a request handler, route, message consumer, or CLI entrypoint.
• Remove a catch with

  finally

  that does cleanup (closing resources, releasing locks).
• Remove a catch that converts one error type to another semantically meaningful one.
• "Fix" Go's

  if err != nil { return err }

  patterns — that IS the proper Go way, not defensive code.
• Remove

  unwrap()

  from Rust without understanding the precondition; replacement should be

  expect("reason")

  minimum, or proper handling.
• Remove a Python

  except

  without checking if it's catching a specific expected exception.
• Modify error handling to "make tests pass" — the test failure may be the actual bug.