galaxy-tool-lint

The advisory check tier (tier 3.5) of the Galaxy tool refactoring framework: read-only IUC best-practice checks that report but never mutate.

Where the codemod (tier 2) and fmt (tier 3) tiers fix a tool (their detect phases report what their fix would change), this tier covers the IUC tool_xml best-practices that are not safely auto-fixable — they need a content or semantic decision (add tests, declare requirements, write help). Each check is a RuleMeta-coded, detect_only=True rule that yields the shared tier-0.5 Violation. The findings are advisory: the galaxy-tool-refactor check command shows them but, by default, does not fail on them (use --strict to gate on them too).

Scope

Depends only on tier 1 (galaxy-tool-source, the parsed ToolDocument and profile helpers) and tier 0.5 (galaxy-tool-refactor-rules, RuleMeta / Violation). It does not depend on the mutating tiers — the app (galaxy-tool-refactor-cli) composes this alongside them.

from galaxy_tool_source.binding import load_tool
from galaxy_tool_lint.detect import detect_violations

for violation in detect_violations(load_tool("my_tool.xml")):
    print(violation.code, violation.message)

Checks

See galaxy_tool_lint.checks and the IUC coverage map at ../docs/iuc_best_practices.md. GTR020.2 (single-quote Cheetah $var) is implemented — it reads <command> text through the read-only command_text.py lexer (quote/directive-aware, across newlines; the lexer lives in tier 1, galaxy_tool_source.command_text, so the GTR020.1 codemod shares it) and reports one advisory per fully-unquoted shell-line $var. Its provable subset is now auto-fixed by the GTR020.1 codemod (docs/decisions.md D8). GTR089.2 (HelpRstResidual) is the advisory half of the GTR089 partition: it reports the invalid <help> reStructuredText the GTR089.1 repair codemod can't safely fix, through the same tier-1 predicate (galaxy_tool_source.rst; docs/decisions.md D31). GTR032 (&&-vs-lone-&) is a real detector since docs/decisions.md D34 (the data-backed D3 deferral ended when the CT3 lexer landed): the lone_amp.py classifier flags only genuine command-joining lone &s.