hone:magic-number-hunt

Magic Number Hunt

What This Skill Does

Scans source files for literal values that lack explanatory names: numeric magic numbers, unexplained string constants, hardcoded URLs, embedded credentials patterns, timeout/retry values, threshold values, and configuration buried inline. Produces a prioritized list of findings with file paths, line numbers, the offending literal, and a suggested named constant or config key.

When To Use

• Weekly scheduled sweep of an entire repository or workspace.
• Before a release to catch newly introduced unnamed literals.
• When onboarding a new codebase to understand hidden configuration.

Do Not Use

• For style, formatting, or linting issues (use a linter).
• For security credential scanning (use a secrets scanner).
• For type-checking or correctness analysis.
• As a replacement for a full code review.

Inputs To Confirm

1. Scope -- which directories or file patterns to scan (default: entire repo, excluding vendored/generated code).
2. Severity threshold -- whether to report all findings or only high-confidence ones (default: all).
3. Language hints -- any languages to prioritize if the repo is polyglot (default: auto-detect).

Instructions

1. Identify the repository root and enumerate source files, excluding vendored directories (vendor/, node_modules/, .git/, third_party/), generated files, and binary assets.
2. For each source file, scan for literal values that meet any of these criteria:
   • Numeric literals other than 0, 1, -1, 2, and common mathematical constants (pi, e) that appear outside of constant declarations or enum definitions.
   • String literals longer than 3 characters that look like URLs, file paths, hostnames, API endpoints, version strings, or configuration keys, and are not already in a constants file or config module.
   • Timeout, retry, and threshold values -- any number used as a duration, count, size limit, or boundary that is not assigned to a named constant.
   • Hardcoded booleans in function calls -- e.g., foo(true, false) where the meaning is unclear without a named parameter.
3. For each finding, record:
   • File path and line number.
   • The literal value.
   • The surrounding code context (2-3 lines).
   • A severity tag: high (likely config leak or unexplained threshold), medium (unnamed numeric/string constant), or low (borderline case, may be intentional).
   • A suggested constant name and placement (e.g., "extract to MAX_RETRY_ATTEMPTS = 3 in config module").
4. Group findings by file, then sort by severity descending within each file.
5. Exclude known false positives:
   • Array indices and loop bounds of 0 or 1.
   • Test fixture data clearly scoped to test files.
   • String literals used only in log messages or error messages (unless they contain URLs or config).
   • Enum or constant definitions themselves.
6. Produce the output report.