Megatron-Core Linting & Formatting

The skill only directs local linting and formatting tools; no malware, credential handling, or exfiltration is evident. It uses uv sync --locked, and the repository CI shows constrained permissions. Deductions apply because fix mode modifies files without explicit user confirmation, backup/rollback guidance, data-flow disclosure, execution isolation, or clear recovery boundaries.

The commands, tool list, and check/fix paths are broadly consistent, and the benchmark reports passing static validation. Deductions apply because autoformat.sh, pyproject.toml, and the relevant CI configuration are not included in the target skill evidence; abnormal inputs, missing tools, version incompatibility, and diagnostic failure handling are unspecified. Under static calibration, the score remains at or below 10.

The audience, Megatron-LM context, trigger phrases, and main tools are reasonably clear, covering pre-PR checks and common lint issues. Deductions apply because non-fit cases, prerequisites, platform differences, and Chinese-language guidance are not stated. The container and uv assumptions limit environment fit, although the core function does not depend on overseas online services.

Front matter supplies the name, description, author, and Apache-2.0 license; the skill card adds use case, output, version, owner, and references, while command examples support progressive use. Deductions apply for missing metadata.tags, Instructions, Examples, and Purpose sections noted by the report, plus limited FAQ/known-limitations coverage, no substantive changelog, and no clear maintenance/update path. Repository license metadata is NOASSERTION.

The skill gives directly reusable commands and style rules for the core Megatron-LM linting and formatting task, likely reducing manual lookup. Deductions apply because no actual outputs, remediation workflow, or result verification are shown, and users must still validate script and tool configuration. The benchmark provides no usable Tier 3 results, so static calibration limits the score to 7 or below.

Commands, tool declarations, and benchmark static findings are reviewable, and the skill card provides a version and reference documentation. Deductions apply because the benchmark explicitly lacks Tier 3 task details, signals, and rollup, evals.json is empty, and no committed CI/test evidence covers this skill's key paths. Evidence is therefore limited to static material and remains at or below 5.