正規表現をテスト表で検証（キャプチャ確認・ReDoS 検出）

What it does

A procedure for never handing over an untested regex.

• Loads a case table in text form (+ must match / - must not match) or JSON and runs the pattern with Python re
• Prints PASS/FAIL, the match, span, and numbered and named groups for each case; JSON cases can also assert expected groups
• Statically flags catastrophic-backtracking (ReDoS) constructs such as nested quantifiers like (a+)+ or (\w+\s?)+ and repeated overlapping alternatives
• Runs every case in a worker process with a timeout so a runaway pattern cannot hang, and --stress times risky fragments on growing input
• Includes checks for common traps such as $ matching before a trailing newline and \d matching full-width digits

How to use

python scripts/regex_test.py examples/postal.txt
python scripts/regex_test.py examples/email.json --stress
python scripts/regex_test.py --pattern "(a+)+$" --lint-only

Ask Claude, Codex, Hermes Agent or Cursor to write or fix a regex; it writes the test table first, iterates until every case passes, and delivers the pattern with the results.

Requirements

• Python 3.10+ (standard library only; atomic groups and possessive quantifiers need 3.11+)

Caveats

• Tests run on Python re. Other engines such as JavaScript or Go differ in syntax and behavior; porting notes are part of the procedure.
• The backtracking lint is heuristic. A clean result is not proof of safety, so keep long non-matching cases in the table.
• It only reads the case file; nothing is modified or sent over the network.