fordaaaa/panoply/polyglot

Use this when a command needs to verify code and the repo's stack is unknown or mixed. Detect from the files and manifests actually present — never assume a runner, a package manager, or a test suite exists.

Detection — manifests beat vibes

Check the touched area for these signals, in order. Several may match at once — that means polyglot, and each area reports its own stack.

• Python: pyproject.toml, setup.cfg, requirements*.txt, uv.lock, .venv
• Go: go.mod
• Rust: Cargo.toml
• Node/TypeScript: package.json (read packageManager + scripts; never assume npm)
• Java: pom.xml, build.gradle*
• Docker: Dockerfile, compose*.yml, docker-compose*.yml (note presence only — never build or start images unasked)

Commands matrix

Prefer the repo's declared script over the raw tool (e.g. scripts.test before bare pytest). Probe cheaply (--version / --help) before running anything long.

Rules

• Read-only first: manifests and scripts before any execution.
• Cheapest probe first: --version before the suite, one package before ./....
• If no runner exists for a stack, report no-framework loudly. An unverifiable claim is not a satisfied one.
• Never touch .github/workflows/, CI config, lockfiles, or secrets to make a check pass.

Report-back format (required)

End any verification block with one line per command run:

[polyglot: <stack> `<cmd>` → exit <n> · <summary line or "none found">]

If no command was run, the line reads [polyglot: <stack> no command — <why>]. This is the artifact that makes the skill checkable: a reader can point at the line and say whether the check ran.