Menu

Individual skill

adversarial-python

0upvotes

Use this skill to gate Python code (floors 3.10+, rules verified through 3.14) with a pass/fail adversarial review — a single blind reviewer subagent judges a diff or file set against 20 decidable rules hunting two failure modes. First, code modern Python makes unnecessary — branch ladders over match/registries, hand-written init/repr/eq over dataclasses, TypeVar ritual over PEP 695, typing.Optional over PEP 604 unions, os.path over pathlib, hand-rolled stdlib batteries, deprecated utcnow, orphan create_task over TaskGroup. Second, legacy-pattern propagation — single-implementation ABCs, pass-through layers, single-method classes, concrete-inheritance reuse, boolean-forked functions, shapeless payloads and parameter clumps — judged as if greenfield; consistency with legacy code is not PASS evidence. A version probe reads the target's Python floor, marks rules above it N/A, and fetches the official what's-new delta when the floor exceeds the verified version. Verdicts only, never fixes.

skills/.experimental/adversarial-python/SKILL.md

Read pinned source

Source commit cf93c57cac89d6fc3e4194686000411567f5caf3

What was checked

Matches the observed repository HEAD
Yes
Scanner rules are current
No
Repository scan complete
No
Skill scan complete
No
Observed behavior
Not checked yet
Last checked
Nov 5, 2026

These are observations at the recorded commit, not a guarantee of safety. Review the source before running it.

Install this skill

Choose an agent in your Stack or copy a command below. Commands fetch the pinned commit and copy only this skill directory. Existing destinations are refused. Copying a command does not install anything.

An install command is unavailable: current rules, complete repository and skill scans, an eligible behavior level and a permitted license are required.

Repository context

Repository stars (not skill ratings)
222
Repository last updated
Aug 15, 2026
License
MIT

Observed agent integration: No specific integration evidence. Standard SKILL.md installation does not imply verified compatibility.

Files and excerpt

File metadata and permitted, redacted excerpts come from the recorded scan. Source links are pinned to that commit.

Browse files
Read excerpt

# Adversarial Python Gate

A modern-idiom and code-structure review gate for Python — pass/fail: a single blind reviewer subagent judges the work against this gate's rules with an adversarial mandate, and the work passes only when every rule is PASS or N/A. This skill renders verdicts; it never fixes the work.

The rules target two failure modes with one root cause — the author reproduced a shape instead of designing one. **Training-data inertia** produces code a modern Python feature deletes outright: the `if/elif` ladder that `match` or a registry replaces, the `__init__`/`__repr__`/`__eq__` triple that `@dataclass(slots=True)` generates, the `TypeVar` ritual PEP 695 retired, the chunking helper `itertools.batched` shipped. **Legacy-pattern propagation** produces new code faithfully extending the surrounding codebase's bad structure — one more branch on the event ladder, one more method on the pass-through service — instead of tracing the feature end-to-end and modeling it. Each rule carries an **Evidence of violation** paragraph so a reviewer can decide PASS/FAIL/N/A from artifact evidence alone, and a `Requires Python ≥ 3.X` gate where the fix depends on a language version.

## When to Apply

- A Python feature, endpoint, or module (agent-authored or human) is about to merge and needs an objective PASS/FAIL on whether modern Python and a fresh architectural look would delete or restructure it.
- An agent extended a legacy codebase and you suspect it copied the existing patterns — event/version branch ladders, service layers that only forward, stringly-typed state — instead of re-architecting the feature.
- A codebase raised its Python floor (to 3.12, 3.13, 3.14+) and changed code should be held to the idioms the new floor enables.
- A refactor claims to modernize or simplify and you want the claim verdict-checked rather than diff-skimmed.

Do not apply to targets with no Python source (the reviewer prompt's precondition aborts with "GATE NOT APPLICABLE"), or when the user wants explanations and refactors rather than a verdict. Judgment calls the gate deliberately excludes — naming taste, function length, docstring and test coverage, performance tuning — belong to advisory skills, not this gate.