Audit Task

$ARGUMENTS

LARP Assessment: Find Fake Code

LLMs are prone to writing "performative" code that looks correct but doesn't actually work. Your job is to find it.

What is Faking?

Code Faking

• •Stub functions that return hardcoded values
• •Copy-paste drift - copied code modified slightly to "work" instead of fixing the real issue
• •Library avoidance - rewriting functionality because the real library was "too hard"
• •Happy path only - code that works for the demo case but crashes on real input

Test Faking

• •Mocking the thing being tested - tests that pass because the real code is mocked out
• •Skipped tests with TODOs that will never be done
• •Tests that assert nothing - they run green but verify nothing
• •Changed tests to match broken behavior instead of fixing the code

Error Faking

• •Silent swallows - catch (e) {} or except: pass
• •Fake retries - retry logic that doesn't actually retry
• •Validation theater - validation functions that return true regardless

Async Faking

• •Fire and forget - async calls without await
• •Promise chains that drop errors
• •Timeouts set to huge values to avoid dealing with timing

Red Flags to Look For

• •# TODO, // FIXME, pass, ..., NotImplementedError
• •Functions under 5 lines that should be complex
• •Tests with lots of mocks and few assertions
• •Error handling that logs but doesn't handle
• •"Temporary" code with no removal date

Output Format

For each finding, include:

• •File:line reference
• •What the fake code is
• •Why it's fake (what would actually happen)
• •Severity (CRITICAL/WARNING/SUGGESTION)

Success Criteria

Finding no fake code is a valid outcome - it means the implementation is genuine. The goal is real, working code, not finding problems.

You are auditing only. Return findings. User will run /0-fix.