moonrunnerkc/skillcheck

Cross-agent skill quality gate for SKILL.md files. Validates against the agentskills.io specification.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Activelast commit Jul 9, 2026
License
MIT

Pinned Snippet

workflow.ymlSHA-pinned
uses: moonrunnerkc/skillcheck@f2c2946e7007cc745a66d365849a1a740b24fb2c # v1.4.1

tags can be moved; commit SHAs can't. why a SHA?

namedescriptionrequireddefault
pathPath to a SKILL.md file or directory to scan recursively.no.
versionskillcheck version to install (e.g., '1.2.3'). Leave empty for latest.no""
min-desc-scoreMinimum description quality score (0-100). Below this triggers a warning.no""
target-agentScope compat checks: claude, vscode, cursor, or all (default: all).no""
strict-vscodePromote VS Code compat issues to errors.nofalse
strict-cursorPromote Cursor compat issues to errors.nofalse
strictStrict mode. Escalates warnings to exit 1 and enables --strict-vscode and --strict-cursor.nofalse
skip-dirname-checkSkip directory-name matching check.nofalse
skip-ref-checkSkip file reference validation.nofalse
ignoreComma-separated rule prefixes to suppress (e.g., 'sizing,disclosure').no""
max-linesOverride the line-count threshold (default: 500).no""
max-tokensOverride the token-count threshold (default: 8000).no""
formatOutput format. Defaults to github for automatic PR annotations.nogithub
semanticEnable semantic-adjacent validation. In standalone CI this enables heuristic graph analysis.nofalse
analyze-graphRun heuristic capability graph analyzers and merge diagnostics into the report.nofalse
ingest-critiquePath to an agent self-critique JSON response to ingest.no""
critique-agentCritique prompt/source agent: claude, codex, or cursor.no""
ingest-graphPath to an agent graph-extraction JSON response to ingest.no""
graph-agentGraph prompt/source agent: claude, codex, or cursor.no""
historyAppend a validation record to .skillcheck-history.json.nofalse
fail-on-regressionWith --history, exit 1 on history.skill.regressed. Independent of --strict.nofalse
explain-scoreShow per-dimension breakdown for description quality scores.nofalse
activation-hypothesesGenerate activation hypotheses. This is an emit mode and does not produce PR annotations.nofalse
tiktokenInstall the optional tiktoken extra for more accurate token estimates.nofalse

no outputs