misleading-name-detector

Misleading name detector: harvests every prefixed function with reader/mutator/predicate cues (get*/find*/fetch*/set*/write*/is*/has*/can*/...) and extracts the first 600 chars of brace-balanced body.

SKILL.md

GitHub

What this is for

getUser() that pushes a default user. isValid() that returns a string. tempFix() from 2019. Identifiers whose names promise something the implementation does not deliver - linguistic anti-patterns the type system cannot see.

The skill harvests every prefixed function with conventional cues - reader (get*, find*, fetch*), mutator (set*, write*, push*), predicate (is*, has*, can*) - and pairs the symbol name with the first 600 chars of brace-balanced body. The LLM judges whether behavior matches promise. The collector never raises false positives alone, but it gives the LLM exactly what it needs to do so.

PROTECTION RULE - never ~/.claude/

Read-only skill. Guard required if write mode added later.

## What You Must Do When Invoked

Step 1

  1. If -help is passed, print the ## Usage block below and stop.

Step 2

  1. Confirm -ProjectDir is provided and the path exists.

Step 3

  1. Run: scripts/name-harvest.ps1 -ProjectDir "<path>"

Step 4

  1. LLM reads the JSON output. For each symbol with prefix-kind + body preview:

Step 5

  • Does the body do what the name promises?

Step 6

  • Readers (get* etc.) must not mutate or perform write I/O.

Step 7

  • Predicates (is*/has*/can* etc.) must return boolean.

Step 8

  • Mutators (set*/write*/push* etc.) are not subject to a wrong-kind

Step 9

finding, but their name should reflect what they mutate.

Step 10

  • Async/Sync suffix must match actual sync vs. promise-yielding behavior.

Step 11

  1. Classify: truthful / lies-kind (severity by API surface and call count).

Step 12

  1. Confidence: proven (direct code evidence), likely (multiple cues),

Step 13

suspected (judgment call).

Step 14

  1. Write name-lies-report.md to the working directory.

Usage

/name-lies                           # interactive, prompts for directory
/name-lies <dir>                     # scan project directory
/name-lies -help                     # show usage

Returns JSON with symbols[]: {file, line, name, prefixKind, mutations[], ios[], isAsync, hasAsyncSuffix, bodyPreview} plus counts summary.

Report Format

name-lies-report.md:

  • Executive summary (total prefixed symbols, % truthful vs lies)
  • Critical findings (promise breaks for widely-called public APIs)
  • Medium findings (less-visible naming lies, fix candidates)
  • Low findings (inconclusive - body too short to judge)
  • Confidence column
  • Open questions (suspected)