skilltest 0.7.0 → 0.8.0

package/dist/index.js

@@ -1,13 +1,13 @@
 #!/usr/bin/env node
 
 // src/index.ts
-import fs11 from "node:fs";
-import path6 from "node:path";
+import fs12 from "node:fs";
+import path7 from "node:path";
 import { fileURLToPath } from "node:url";
 import { Command } from "commander";
 
 // src/commands/lint.ts
-import fs6 from "node:fs/promises";
+import fs7 from "node:fs/promises";
 import { z as z6 } from "zod";
 
 // src/core/skill-parser.ts
@@ -972,6 +972,116 @@ function runFrontmatterChecks(context) {
   return issues;
 }
 
+// src/core/linter/plugin.ts
+import fs4 from "node:fs/promises";
+import path4 from "node:path";
+import { pathToFileURL } from "node:url";
+function normalizeRuleCheckId(checkId) {
+  return checkId.includes(":") ? checkId : `plugin:${checkId}`;
+}
+function buildPluginValidationError(filePath, message) {
+  return new Error(`Invalid lint plugin at ${filePath}: ${message}`);
+}
+function validatePluginCandidate(candidate, filePath, exportName) {
+  if (!candidate || typeof candidate !== "object" || !("rules" in candidate)) {
+    throw buildPluginValidationError(filePath, `${exportName} export must be an object with a rules array.`);
+  }
+  const rules = candidate.rules;
+  if (!Array.isArray(rules)) {
+    throw buildPluginValidationError(filePath, `${exportName} export must include a rules array.`);
+  }
+  return {
+    rules: rules.map((rule, index) => {
+      if (!rule || typeof rule !== "object") {
+        throw buildPluginValidationError(filePath, `rule at index ${index} must be an object.`);
+      }
+      const checkId = rule.checkId;
+      if (typeof checkId !== "string" || checkId.trim() === "") {
+        throw buildPluginValidationError(filePath, `rule at index ${index} must have a non-empty string checkId.`);
+      }
+      const title = rule.title;
+      if (typeof title !== "string" || title.trim() === "") {
+        throw buildPluginValidationError(filePath, `rule at index ${index} must have a non-empty string title.`);
+      }
+      const check = rule.check;
+      if (typeof check !== "function") {
+        throw buildPluginValidationError(filePath, `rule '${checkId}' must have a check function.`);
+      }
+      return {
+        checkId: normalizeRuleCheckId(checkId),
+        title,
+        check
+      };
+    })
+  };
+}
+async function loadPlugin(filePath) {
+  const absolutePath = path4.resolve(filePath);
+  try {
+    await fs4.access(absolutePath);
+  } catch {
+    throw new Error(`Failed to load lint plugin at ${absolutePath}: file does not exist.`);
+  }
+  let loadedModule;
+  try {
+    loadedModule = await import(pathToFileURL(absolutePath).href);
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    throw new Error(`Failed to load lint plugin at ${absolutePath}: ${message}`);
+  }
+  const validationErrors = [];
+  for (const [exportName, candidate] of [
+    ["default", loadedModule.default],
+    ["plugin", loadedModule.plugin]
+  ]) {
+    if (candidate === void 0) {
+      continue;
+    }
+    try {
+      return validatePluginCandidate(candidate, absolutePath, exportName);
+    } catch (error) {
+      validationErrors.push(error instanceof Error ? error.message : String(error));
+    }
+  }
+  if (validationErrors.length > 0) {
+    throw new Error(validationErrors.join(" "));
+  }
+  throw buildPluginValidationError(
+    absolutePath,
+    "expected a default export or named export 'plugin' containing a rules array."
+  );
+}
+function buildRuleExecutionError(rule, error) {
+  const message = error instanceof Error ? error.message : String(error);
+  return {
+    id: `plugin.load-error.${rule.checkId.replace(/[^A-Za-z0-9]+/g, "-").replace(/^-+|-+$/g, "").toLowerCase()}`,
+    checkId: "plugin:load-error",
+    title: "Plugin Rule Error",
+    status: "fail",
+    message: `Plugin rule '${rule.checkId}' failed: ${message}`
+  };
+}
+async function runPluginRules(plugin, context) {
+  const issues = [];
+  for (const rule of plugin.rules) {
+    try {
+      const result = await rule.check(context);
+      if (!Array.isArray(result)) {
+        throw new Error("check function must return an array of lint issues.");
+      }
+      issues.push(
+        ...result.map((issue) => ({
+          ...issue,
+          checkId: rule.checkId
+        }))
+      );
+    } catch (error) {
+      issues.push(buildRuleExecutionError(rule, error));
+    }
+  }
+  return issues;
+}
+
 // src/core/linter/security.ts
 var DANGEROUS_COMMAND_PATTERNS = [
   {
@@ -1179,8 +1289,8 @@ function runSecurityChecks(context) {
 }
 
 // src/core/linter/structure.ts
-import fs4 from "node:fs/promises";
-import path4 from "node:path";
+import fs5 from "node:fs/promises";
+import path5 from "node:path";
 function hasTableOfContents(content) {
   if (/^#{1,6}\s+table of contents\b/im.test(content)) {
     return true;
@@ -1221,21 +1331,21 @@ async function runStructureChecks(context) {
       message: `SKILL.md length is ${context.skill.lineCount} lines.`
     });
   }
-  const referencesDir = path4.join(context.skill.skillRoot, "references");
+  const referencesDir = path5.join(context.skill.skillRoot, "references");
   if (await pathExists(referencesDir)) {
     const files = await listFilesRecursive(referencesDir);
     let oversizedWithoutToc = 0;
     for (const file of files) {
-      const raw = await fs4.readFile(file, "utf8");
+      const raw = await fs5.readFile(file, "utf8");
       const lineCount = raw === "" ? 0 : raw.split(/\r?\n/).length;
       if (lineCount > 300 && !hasTableOfContents(raw)) {
         oversizedWithoutToc += 1;
         issues.push({
-          id: `structure.references.toc.${toPosixPath(path4.relative(context.skill.skillRoot, file))}`,
+          id: `structure.references.toc.${toPosixPath(path5.relative(context.skill.skillRoot, file))}`,
           checkId: "structure:toc",
           title: "Reference File Navigation",
           status: "warn",
-          message: `${toPosixPath(path4.relative(context.skill.skillRoot, file))} is ${lineCount} lines and has no table of contents.`,
+          message: `${toPosixPath(path5.relative(context.skill.skillRoot, file))} is ${lineCount} lines and has no table of contents.`,
           suggestion: "Add a table of contents for long reference files."
         });
       }
@@ -1265,7 +1375,7 @@ async function runStructureChecks(context) {
     other: []
   };
   for (const reference of references) {
-    const resolved = path4.resolve(context.skill.skillRoot, reference);
+    const resolved = path5.resolve(context.skill.skillRoot, reference);
     if (!await pathExists(resolved)) {
       const kind = classifyReferencePath(reference);
       missingByType[kind].push(reference);
@@ -1362,6 +1472,10 @@ async function runLinter(inputPath, options = {}) {
   issues.push(...runSecurityChecks(context));
   issues.push(...await runDisclosureChecks(context));
   issues.push(...runCompatibilityChecks(context));
+  for (const pluginPath of options.plugins ?? []) {
+    const plugin = await loadPlugin(pluginPath);
+    issues.push(...await runPluginRules(plugin, context));
+  }
   const filteredIssues = issues.filter((issue) => !suppressedCheckIds.has(issue.checkId));
   return {
     target: inputPath,
@@ -1525,10 +1639,10 @@ function renderLintIssueList(report) {
   const info = skippedSecurityPatterns > 0 ? `<p class="info-line">Skipped security patterns in examples/comments: ${escapeHtml(skippedSecurityPatterns)}</p>` : "";
   return `<div class="row-list">${rows}</div>${info}`;
 }
-function renderTriggerCaseRow(testCase) {
+function renderTriggerCaseRow(testCase, showSelectedCompetitor) {
   const details = testCase.rawModelResponse ? renderDetails("Model response", renderPreBlock(testCase.rawModelResponse)) : "";
   return `
-    <div class="row">
+    <div class="row${testCase.selectedCompetitor ? " competitor-selected" : ""}">
       <div class="row-header">
         <div>
           <div class="row-title">${escapeHtml(testCase.query)}</div>
@@ -1540,12 +1654,29 @@ function renderTriggerCaseRow(testCase) {
       </div>
       ${renderDefinitionList([
     { label: "Expected", value: testCase.expected },
-    { label: "Actual", value: testCase.actual }
+    { label: "Actual", value: testCase.actual },
+    ...showSelectedCompetitor ? [{ label: "Selected competitor", value: testCase.selectedCompetitor ?? "none" }] : []
   ])}
       ${details}
     </div>
   `;
 }
+function renderCompetitorSkillsSection(result) {
+  if (!result.competitors || result.competitors.length === 0) {
+    return "";
+  }
+  return renderSectionCard(
+    "Competitor Skills",
+    `<div class="row-list">${result.competitors.map(
+      (competitor) => renderMessageRow(
+        "warn",
+        competitor.name,
+        competitor.description,
+        renderDefinitionList([{ label: "Source", value: competitor.sourcePath }])
+      )
+    ).join("")}</div>`
+  );
+}
 function promptStatus(promptResult) {
   if (promptResult.totalAssertions === 0) {
     return "skip";
@@ -1638,6 +1769,7 @@ function renderHtmlDocument(title, body) {
         --pass: #22c55e;
         --warn: #eab308;
         --fail: #ef4444;
+        --competitor: #f97316;
         --skip: #6b7280;
         --shadow: 0 10px 30px rgba(15, 23, 42, 0.08);
       }
@@ -1786,6 +1918,11 @@ function renderHtmlDocument(title, body) {
         background: var(--surface-muted);
       }
 
+      .row.competitor-selected {
+        border-color: rgba(249, 115, 22, 0.45);
+        background: rgba(249, 115, 22, 0.08);
+      }
+
       .row-header {
         display: flex;
         justify-content: space-between;
@@ -1965,6 +2102,7 @@ function renderTriggerHtml(result) {
   const target = resolveOptionalTarget(htmlResult, result.skillName);
   const matchedCount = result.cases.filter((testCase) => testCase.matched).length;
   const matchRate = result.cases.length === 0 ? 0 : matchedCount / result.cases.length;
+  const hasCompetitors = Boolean(result.competitors && result.competitors.length > 0);
   const body = [
     renderHeaderCard(
       "trigger",
@@ -1980,10 +2118,15 @@ function renderTriggerHtml(result) {
         { label: "Provider", value: result.provider },
         { label: "Model", value: result.model },
         { label: "Seed", value: result.seed !== void 0 ? String(result.seed) : "none" },
+        ...hasCompetitors ? [{ label: "Competitors", value: String(result.competitors?.length ?? 0) }] : [],
         { label: "Queries", value: String(result.queries.length) }
       ]
     ),
-    renderSectionCard("Trigger Cases", `<div class="row-list">${result.cases.map((testCase) => renderTriggerCaseRow(testCase)).join("")}</div>`),
+    renderCompetitorSkillsSection(result),
+    renderSectionCard(
+      "Trigger Cases",
+      `<div class="row-list">${result.cases.map((testCase) => renderTriggerCaseRow(testCase, hasCompetitors)).join("")}</div>`
+    ),
     renderSectionCard(
       "Suggestions",
       `<ul>${result.suggestions.map((suggestion) => `<li>${escapeHtml(suggestion)}</li>`).join("")}</ul>`
@@ -2023,7 +2166,8 @@ function renderEvalHtml(result) {
 }
 function renderCheckHtml(result) {
   const skillName = result.trigger?.skillName ?? result.eval?.skillName ?? result.target;
-  const triggerBody = result.trigger ? `<div class="row-list">${result.trigger.cases.map((testCase) => renderTriggerCaseRow(testCase)).join("")}</div>
+  const triggerBody = result.trigger ? `${renderCompetitorSkillsSection(result.trigger)}
+       <div class="row-list">${result.trigger.cases.map((testCase) => renderTriggerCaseRow(testCase, Boolean(result.trigger?.competitors?.length))).join("")}</div>
        <div class="card" style="margin-top: 16px;">
          <h2>Trigger Suggestions</h2>
          <ul>${result.trigger.suggestions.map((suggestion) => `<li>${escapeHtml(suggestion)}</li>`).join("")}</ul>
@@ -2123,46 +2267,47 @@ function countSkippedSecurityPatterns2(issues) {
     return total + (issue.skippedPatterns?.length ?? 0);
   }, 0);
 }
+function formatPercent2(value) {
+  return `${(value * 100).toFixed(1)}%`;
+}
 function renderLintReport(report, enableColor) {
   const c = getChalkInstance(enableColor);
   const { passed, warnings, failures, total } = report.summary;
   const headerLines = [
-    `\u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510`,
-    `\u2502 skilltest lint                                                \u2502`,
-    `\u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524`,
-    `\u2502 target: ${report.target}`,
-    `\u2502 summary: ${passed}/${total} checks passed, ${warnings} warnings, ${failures} failures`,
-    `\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518`
+    "skilltest lint",
+    `target: ${report.target}`,
+    `summary: ${passed}/${total} checks passed, ${warnings} warnings, ${failures} failures`
   ];
   const renderedIssues = report.issues.map((issue) => renderIssueLine(issue, c)).join("\n");
   const skippedSecurityPatterns = countSkippedSecurityPatterns2(report.issues);
   const infoLine = skippedSecurityPatterns > 0 ? `
-  ${c.cyan("\u2139")} ${skippedSecurityPatterns} security pattern(s) found in code examples/comments (not flagged)` : "";
+  ${c.cyan("INFO")} ${skippedSecurityPatterns} security pattern(s) found in code examples/comments (not flagged)` : "";
   return `${headerLines.join("\n")}
 ${renderedIssues}${infoLine}`;
 }
-function formatPercent2(value) {
-  return `${(value * 100).toFixed(1)}%`;
-}
 function renderTriggerReport(result, enableColor, verbose) {
   const c = getChalkInstance(enableColor);
-  const lines = [];
-  lines.push("\u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510");
-  lines.push("\u2502 skilltest trigger                                             \u2502");
-  lines.push("\u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524");
-  lines.push(`\u2502 skill: ${result.skillName}`);
-  lines.push(`\u2502 provider/model: ${result.provider}/${result.model}`);
+  const lines = [
+    "skilltest trigger",
+    `skill: ${result.skillName}`,
+    `provider/model: ${result.provider}/${result.model}`
+  ];
+  if (result.competitors && result.competitors.length > 0) {
+    lines.push(`competitors: ${result.competitors.map((competitor) => competitor.name).join(", ")}`);
+  }
   lines.push(
-    `\u2502 precision: ${formatPercent2(result.metrics.precision)}  recall: ${formatPercent2(result.metrics.recall)}  f1: ${formatPercent2(result.metrics.f1)}`
+    `precision: ${formatPercent2(result.metrics.precision)}  recall: ${formatPercent2(result.metrics.recall)}  f1: ${formatPercent2(result.metrics.f1)}`
   );
   lines.push(
-    `\u2502 TP ${result.metrics.truePositives}  TN ${result.metrics.trueNegatives}  FP ${result.metrics.falsePositives}  FN ${result.metrics.falseNegatives}`
+    `TP ${result.metrics.truePositives}  TN ${result.metrics.trueNegatives}  FP ${result.metrics.falsePositives}  FN ${result.metrics.falseNegatives}`
   );
-  lines.push("\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518");
   for (const [index, testCase] of result.cases.entries()) {
     const status = testCase.matched ? c.green("PASS") : c.red("FAIL");
     lines.push(`${index + 1}. ${status} query: ${testCase.query}`);
     lines.push(`   expected: ${testCase.expected} | actual: ${testCase.actual}`);
+    if (verbose && testCase.selectedCompetitor) {
+      lines.push(`   competitor selected: ${testCase.selectedCompetitor}`);
+    }
     if (verbose && testCase.rawModelResponse) {
       lines.push(`   model: ${testCase.rawModelResponse.replace(/\s+/g, " ").trim()}`);
     }
@@ -2175,15 +2320,13 @@ function renderTriggerReport(result, enableColor, verbose) {
 }
 function renderEvalReport(result, enableColor, verbose) {
   const c = getChalkInstance(enableColor);
-  const lines = [];
-  lines.push("\u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510");
-  lines.push("\u2502 skilltest eval                                                \u2502");
-  lines.push("\u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524");
-  lines.push(`\u2502 skill: ${result.skillName}`);
-  lines.push(`\u2502 provider/model: ${result.provider}/${result.model}`);
-  lines.push(`\u2502 grader model: ${result.graderModel}`);
-  lines.push(`\u2502 assertions passed: ${result.summary.passedAssertions}/${result.summary.totalAssertions}`);
-  lines.push("\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518");
+  const lines = [
+    "skilltest eval",
+    `skill: ${result.skillName}`,
+    `provider/model: ${result.provider}/${result.model}`,
+    `grader model: ${result.graderModel}`,
+    `assertions passed: ${result.summary.passedAssertions}/${result.summary.totalAssertions}`
+  ];
   for (const [index, promptResult] of result.results.entries()) {
     lines.push(`${index + 1}. prompt: ${promptResult.prompt}`);
     lines.push(`   response summary: ${promptResult.responseSummary.replace(/\s+/g, " ").trim()}`);
@@ -2229,7 +2372,7 @@ function renderCheckReport(result, enableColor, verbose) {
   }
   const skippedSecurityPatterns = countSkippedSecurityPatterns2(result.lint.issues);
   if (skippedSecurityPatterns > 0) {
-    lines.push(`  ${c.cyan("\u2139")} ${skippedSecurityPatterns} security pattern(s) found in code examples/comments (not flagged)`);
+    lines.push(`  ${c.cyan("INFO")} ${skippedSecurityPatterns} security pattern(s) found in code examples/comments (not flagged)`);
   }
   lines.push("");
   lines.push("Trigger");
@@ -2240,11 +2383,17 @@ function renderCheckReport(result, enableColor, verbose) {
     lines.push(
       `  TP ${result.trigger.metrics.truePositives} TN ${result.trigger.metrics.trueNegatives} FP ${result.trigger.metrics.falsePositives} FN ${result.trigger.metrics.falseNegatives}`
     );
+    if (result.trigger.competitors && result.trigger.competitors.length > 0) {
+      lines.push(`  competitors: ${result.trigger.competitors.map((competitor) => competitor.name).join(", ")}`);
+    }
     const triggerCases = verbose ? result.trigger.cases : result.trigger.cases.filter((testCase) => !testCase.matched);
     for (const testCase of triggerCases) {
       const status = testCase.matched ? c.green("PASS") : c.red("FAIL");
       lines.push(`  - ${status} ${testCase.query}`);
       lines.push(`    expected=${testCase.expected} actual=${testCase.actual}`);
+      if (testCase.selectedCompetitor) {
+        lines.push(`    competitor selected=${testCase.selectedCompetitor}`);
+      }
     }
   } else {
     lines.push(`- ${triggerGate} ${result.triggerSkippedReason ?? "Skipped."}`);
@@ -2286,7 +2435,7 @@ function renderCheckReport(result, enableColor, verbose) {
 }
 
 // src/commands/common.ts
-import fs5 from "node:fs/promises";
+import fs6 from "node:fs/promises";
 import { z as z5 } from "zod";
 
 // src/core/eval-runner.ts
@@ -2568,7 +2717,7 @@ function parseJsonArrayFromModelOutput(raw) {
   }
   throw new Error("Model did not return a JSON array.");
 }
-async function generateQueriesWithModel(skill, provider, model, numQueries) {
+async function generateQueriesWithModel(skill, provider, model, numQueries, competitors) {
   validateNumQueries(numQueries);
   const shouldTriggerCount = Math.floor(numQueries / 2);
   const shouldNotTriggerCount = numQueries - shouldTriggerCount;
@@ -2581,6 +2730,15 @@ async function generateQueriesWithModel(skill, provider, model, numQueries) {
   const userPrompt = [
     `Skill name: ${skill.frontmatter.name}`,
     `Skill description: ${skill.frontmatter.description}`,
+    ...competitors && competitors.length > 0 ? [
+      "",
+      "Competitor skills in the same domain:",
+      ...competitors.map((competitor) => `- ${competitor.name}: ${competitor.description}`),
+      "",
+      "Generate queries that test whether the target skill triggers correctly even when these similar skills exist.",
+      "Positive queries should clearly belong to the target skill, not the competitors.",
+      "Negative queries should belong to a competitor or to no skill at all."
+    ] : [],
     `Generate ${numQueries} prompts total.`,
     `Exactly ${shouldTriggerCount} should have should_trigger=true.`,
     `Exactly ${shouldNotTriggerCount} should have should_trigger=false.`,
@@ -2614,12 +2772,29 @@ function parseDecision(rawResponse, skillNames) {
   }
   return "unrecognized";
 }
-function prepareTriggerQueries(skill, queries, seed) {
+function prepareTriggerQueries(skill, queries, seed, competitors) {
   const rng = createRng(seed);
+  const competitorCandidates = (competitors ?? []).map((competitor) => ({
+    name: competitor.name,
+    description: competitor.description
+  }));
   return queries.map((testQuery) => {
-    const fakeCount = 5 + Math.floor(rng() * 5);
+    const usingCompetitors = competitorCandidates.length > 0;
+    const fakeCount = usingCompetitors ? testQuery.should_trigger ? 2 + Math.floor(rng() * 3) : 3 + Math.floor(rng() * 3) : 5 + Math.floor(rng() * 5);
     const fakeSkills = sample(FAKE_SKILLS, fakeCount, rng);
-    const allSkills = shuffle(
+    const allSkills = usingCompetitors ? shuffle(
+      [
+        ...competitorCandidates,
+        ...fakeSkills,
+        ...testQuery.should_trigger ? [
+          {
+            name: skill.frontmatter.name,
+            description: skill.frontmatter.description
+          }
+        ] : []
+      ],
+      rng
+    ) : shuffle(
       [
         ...fakeSkills,
         {
@@ -2673,25 +2848,82 @@ function calculateMetrics(skillName, cases) {
     f1
   };
 }
-function buildSuggestions(metrics) {
+function assertCompetitorNamesDistinct(skillName, competitors) {
+  for (const competitor of competitors) {
+    if (competitor.name === skillName) {
+      throw new Error(`Competitor skill '${competitor.name}' has the same name as the skill under test.`);
+    }
+  }
+}
+function buildTriggerCaseResult(options) {
+  const expected = options.testQuery.should_trigger ? options.skillName : "none";
+  const matched = options.testQuery.should_trigger ? options.decision === options.skillName : options.decision !== options.skillName;
+  const selectedCompetitor = options.competitorNames?.includes(options.decision) ? options.decision : void 0;
+  return {
+    query: options.testQuery.query,
+    shouldTrigger: options.testQuery.should_trigger,
+    expected,
+    actual: options.decision,
+    matched,
+    selectedCompetitor,
+    rawModelResponse: options.rawModelResponse
+  };
+}
+function buildSuggestions(skillName, metrics, cases, competitors) {
   const suggestions = [];
   if (metrics.falseNegatives > 0) {
     suggestions.push(
       "False negatives found: clarify capability keywords and add explicit 'use when ...' phrasing in description."
     );
+    if (competitors && competitors.length > 0) {
+      const competitorCounts = /* @__PURE__ */ new Map();
+      for (const testCase of cases) {
+        if (!testCase.shouldTrigger || testCase.actual === skillName || !testCase.selectedCompetitor) {
+          continue;
+        }
+        competitorCounts.set(testCase.selectedCompetitor, (competitorCounts.get(testCase.selectedCompetitor) ?? 0) + 1);
+      }
+      for (const [competitorName, count] of competitorCounts.entries()) {
+        suggestions.push(
+          `Skill '${competitorName}' was selected instead of '${skillName}' for ${count} quer${count === 1 ? "y" : "ies"}. Differentiate your description from '${competitorName}'.`
+        );
+      }
+    }
   }
   if (metrics.falsePositives > 0) {
     suggestions.push("False positives found: narrow scope boundaries and add explicit non-goals in description.");
+    if (competitors && competitors.length > 0) {
+      suggestions.push(
+        `With competitor skills present, ${metrics.falsePositives} negative quer${metrics.falsePositives === 1 ? "y still" : "ies still"} triggered '${skillName}'. Narrow your description's scope boundaries.`
+      );
+    }
   }
   if (suggestions.length === 0) {
     suggestions.push("Trigger behavior looks clean on this sample. Keep monitoring with domain-specific custom queries.");
   }
   return suggestions;
 }
+async function loadCompetitorSkills(comparePaths) {
+  const competitors = [];
+  for (const comparePath of comparePaths) {
+    const parsed = await parseSkillStrict(comparePath);
+    competitors.push({
+      name: parsed.frontmatter.name,
+      description: parsed.frontmatter.description,
+      sourcePath: comparePath
+    });
+  }
+  return competitors;
+}
 async function runTriggerTest(skill, options) {
-  const queries = options.queries && options.queries.length > 0 ? triggerQueryArraySchema.parse(options.queries) : await generateQueriesWithModel(skill, options.provider, options.model, options.numQueries);
+  const competitors = options.compare && options.compare.length > 0 ? await loadCompetitorSkills(options.compare) : void 0;
+  if (competitors && competitors.length > 0) {
+    assertCompetitorNamesDistinct(skill.frontmatter.name, competitors);
+  }
+  const queries = options.queries && options.queries.length > 0 ? triggerQueryArraySchema.parse(options.queries) : await generateQueriesWithModel(skill, options.provider, options.model, options.numQueries, competitors);
   const skillName = skill.frontmatter.name;
-  const preparedQueries = prepareTriggerQueries(skill, queries, options.seed);
+  const preparedQueries = prepareTriggerQueries(skill, queries, options.seed, competitors);
+  const competitorNames = competitors?.map((competitor) => competitor.name) ?? [];
   const systemPrompt = [
     "You are selecting one skill to activate for a user query.",
     "Choose the single best matching skill name from the provided list, or 'none' if no skill is a good fit.",
@@ -2704,18 +2936,15 @@ async function runTriggerTest(skill, options) {
       const rawResponse = await options.provider.sendMessage(systemPrompt, userPrompt, { model: options.model });
       const decision = parseDecision(
         rawResponse,
-        allSkills.map((entry) => entry.name)
+        Array.from(/* @__PURE__ */ new Set([skillName, ...allSkills.map((entry) => entry.name)]))
       );
-      const expected = testQuery.should_trigger ? skillName : "none";
-      const matched = testQuery.should_trigger ? decision === skillName : decision !== skillName;
-      return {
-        query: testQuery.query,
-        shouldTrigger: testQuery.should_trigger,
-        expected,
-        actual: decision,
-        matched,
+      return buildTriggerCaseResult({
+        testQuery,
+        skillName,
+        decision,
+        competitorNames,
         rawModelResponse: options.verbose ? rawResponse : void 0
-      };
+      });
     },
     options.concurrency ?? 5
   );
@@ -2725,10 +2954,11 @@ async function runTriggerTest(skill, options) {
     model: options.model,
     provider: options.provider.name,
     seed: options.seed,
+    competitors,
     queries,
     cases: results,
     metrics,
-    suggestions: buildSuggestions(metrics)
+    suggestions: buildSuggestions(skillName, metrics, results, competitors)
   };
 }
 
@@ -2843,10 +3073,10 @@ async function loadConfiguredEvalPrompts(command) {
   if (!promptFile && assertionsFile) {
     throw new Error("Config field eval.assertionsFile requires eval.promptFile.");
   }
-  const promptRaw = await fs5.readFile(promptFile, "utf8");
+  const promptRaw = await fs6.readFile(promptFile, "utf8");
   let prompts = normalizeEvalPrompts(parseJsonIfPossible(promptRaw), promptFile);
   if (assertionsFile) {
-    const assertionsRaw = await fs5.readFile(assertionsFile, "utf8");
+    const assertionsRaw = await fs6.readFile(assertionsFile, "utf8");
     const assertions = normalizeAssertions(parseJsonIfPossible(assertionsRaw), assertionsFile);
     prompts = prompts.map((prompt) => ({
       prompt: prompt.prompt,
@@ -2885,18 +3115,22 @@ function writeError(error, asJson) {
 
 // src/commands/lint.ts
 var lintCliSchema = z6.object({
-  html: z6.string().optional()
+  html: z6.string().optional(),
+  plugin: z6.array(z6.string().min(1)).optional()
 });
+function collectPluginPaths(value, previous = []) {
+  return [...previous, value];
+}
 async function handleLintCommand(targetPath, options) {
   try {
-    const report = await runLinter(targetPath, { suppress: options.suppress });
+    const report = await runLinter(targetPath, { suppress: options.suppress, plugins: options.plugins });
     if (options.json) {
       writeResult(report, true);
     } else {
       writeResult(renderLintReport(report, options.color), false);
     }
     if (options.html) {
-      await fs6.writeFile(options.html, renderLintHtml(report), "utf8");
+      await fs7.writeFile(options.html, renderLintHtml(report), "utf8");
     }
     if (lintFails(report, options.failOn)) {
       process.exitCode = 1;
@@ -2907,7 +3141,7 @@ async function handleLintCommand(targetPath, options) {
   }
 }
 function registerLintCommand(program) {
-  program.command("lint").description("Run static lint checks against a SKILL.md file or skill directory.").argument("<path-to-skill>", "Path to SKILL.md or skill directory").option("--html <path>", "Write an HTML report to the given file path").action(async (targetPath, _commandOptions, command) => {
+  program.command("lint").description("Run static lint checks against a SKILL.md file or skill directory.").argument("<path-to-skill>", "Path to SKILL.md or skill directory").option("--html <path>", "Write an HTML report to the given file path").option("--plugin <path>", "Load a custom lint plugin file", collectPluginPaths, []).action(async (targetPath, _commandOptions, command) => {
     const globalOptions = getGlobalCliOptions(command);
     const config = getResolvedConfig(command);
     const parsedCli = lintCliSchema.safeParse(command.opts());
@@ -2920,30 +3154,33 @@ function registerLintCommand(program) {
       ...globalOptions,
       failOn: config.lint.failOn,
       suppress: config.lint.suppress,
+      plugins: config.lint.plugins,
       html: parsedCli.data.html
     });
   });
 }
 
 // src/commands/trigger.ts
-import fs8 from "node:fs/promises";
+import fs9 from "node:fs/promises";
 import ora from "ora";
 import { z as z8 } from "zod";
 
 // src/utils/config.ts
-import fs7 from "node:fs/promises";
-import path5 from "node:path";
+import fs8 from "node:fs/promises";
+import path6 from "node:path";
 import { z as z7 } from "zod";
 var providerNameSchema = z7.enum(["anthropic", "openai"]);
 var lintFailOnSchema = z7.enum(["error", "warn"]);
 var lintConfigSchema = z7.object({
   failOn: lintFailOnSchema.optional(),
-  suppress: z7.array(z7.string().min(1)).optional()
+  suppress: z7.array(z7.string().min(1)).optional(),
+  plugins: z7.array(z7.string().min(1)).optional()
 }).strict();
 var triggerConfigSchema = z7.object({
   numQueries: z7.number().int().min(2).refine((value) => value % 2 === 0, "trigger.numQueries must be an even number."),
   threshold: z7.number().min(0).max(1).optional(),
-  seed: z7.number().int().optional()
+  seed: z7.number().int().optional(),
+  compare: z7.array(z7.string().min(1)).optional()
 }).strict().partial();
 var evalConfigSchema = z7.object({
   numRuns: z7.number().int().min(1).optional(),
@@ -2967,12 +3204,14 @@ var resolvedSkilltestConfigSchema = z7.object({
   concurrency: z7.number().int().min(1),
   lint: z7.object({
     failOn: lintFailOnSchema,
-    suppress: z7.array(z7.string().min(1))
+    suppress: z7.array(z7.string().min(1)),
+    plugins: z7.array(z7.string().min(1))
   }),
   trigger: z7.object({
     numQueries: z7.number().int().min(2).refine((value) => value % 2 === 0, "trigger.numQueries must be an even number."),
     threshold: z7.number().min(0).max(1),
-    seed: z7.number().int().optional()
+    seed: z7.number().int().optional(),
+    compare: z7.array(z7.string().min(1))
   }),
   eval: z7.object({
     numRuns: z7.number().int().min(1),
@@ -2988,11 +3227,13 @@ var DEFAULT_SKILLTEST_CONFIG = {
   concurrency: 5,
   lint: {
     failOn: "error",
-    suppress: []
+    suppress: [],
+    plugins: []
   },
   trigger: {
     numQueries: 20,
-    threshold: 0.8
+    threshold: 0.8,
+    compare: []
   },
   eval: {
     numRuns: 5,
@@ -3014,7 +3255,7 @@ function buildConfigValidationError(error, sourceLabel) {
 async function readJsonObject(filePath, label) {
   let raw;
   try {
-    raw = await fs7.readFile(filePath, "utf8");
+    raw = await fs8.readFile(filePath, "utf8");
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error);
     throw new Error(`Failed to read ${label}: ${message}`);
@@ -3038,13 +3279,13 @@ async function loadConfigFromJsonFile(filePath) {
   return {
     configFile: parsed.data,
     sourcePath: filePath,
-    sourceDirectory: path5.dirname(filePath)
+    sourceDirectory: path6.dirname(filePath)
   };
 }
 async function loadConfigFromNearestPackageJson(startDirectory) {
-  let currentDirectory = path5.resolve(startDirectory);
+  let currentDirectory = path6.resolve(startDirectory);
   while (true) {
-    const packageJsonPath = path5.join(currentDirectory, "package.json");
+    const packageJsonPath = path6.join(currentDirectory, "package.json");
     if (await pathExists(packageJsonPath)) {
       const raw = await readJsonObject(packageJsonPath, packageJsonPath);
       const packageJsonSchema = z7.object({
@@ -3063,7 +3304,7 @@ async function loadConfigFromNearestPackageJson(startDirectory) {
         sourceDirectory: currentDirectory
       };
     }
-    const parentDirectory = path5.dirname(currentDirectory);
+    const parentDirectory = path6.dirname(currentDirectory);
     if (parentDirectory === currentDirectory) {
       return null;
     }
@@ -3076,7 +3317,7 @@ async function resolveSkillDirectoryConfig(targetPath) {
   }
   try {
     const { skillRoot } = await resolveSkillPath(targetPath);
-    return loadConfigFromJsonFile(path5.join(skillRoot, ".skilltestrc"));
+    return loadConfigFromJsonFile(path6.join(skillRoot, ".skilltestrc"));
   } catch {
     return null;
   }
@@ -3085,7 +3326,13 @@ function resolveConfigRelativePath(baseDirectory, value) {
   if (!value) {
     return void 0;
   }
-  return path5.resolve(baseDirectory, value);
+  return path6.resolve(baseDirectory, value);
+}
+function resolveConfigRelativePaths(baseDirectory, values) {
+  if (!values || values.length === 0) {
+    return [];
+  }
+  return values.map((value) => path6.resolve(baseDirectory, value));
 }
 function mergeConfigLayers(configFile = {}, cliFlags = {}, baseDirectory = process.cwd()) {
   const merged = {
@@ -3095,12 +3342,20 @@ function mergeConfigLayers(configFile = {}, cliFlags = {}, baseDirectory = proce
     concurrency: cliFlags.concurrency ?? configFile.concurrency ?? DEFAULT_SKILLTEST_CONFIG.concurrency,
     lint: {
       failOn: cliFlags.lint?.failOn ?? configFile.lint?.failOn ?? DEFAULT_SKILLTEST_CONFIG.lint.failOn,
-      suppress: cliFlags.lint?.suppress ?? configFile.lint?.suppress ?? DEFAULT_SKILLTEST_CONFIG.lint.suppress
+      suppress: cliFlags.lint?.suppress ?? configFile.lint?.suppress ?? DEFAULT_SKILLTEST_CONFIG.lint.suppress,
+      plugins: resolveConfigRelativePaths(
+        baseDirectory,
+        cliFlags.lint?.plugins ?? configFile.lint?.plugins ?? DEFAULT_SKILLTEST_CONFIG.lint.plugins
+      )
     },
     trigger: {
       numQueries: cliFlags.trigger?.numQueries ?? configFile.trigger?.numQueries ?? DEFAULT_SKILLTEST_CONFIG.trigger.numQueries,
       threshold: cliFlags.trigger?.threshold ?? configFile.trigger?.threshold ?? DEFAULT_SKILLTEST_CONFIG.trigger.threshold,
-      seed: cliFlags.trigger?.seed ?? configFile.trigger?.seed
+      seed: cliFlags.trigger?.seed ?? configFile.trigger?.seed,
+      compare: resolveConfigRelativePaths(
+        baseDirectory,
+        cliFlags.trigger?.compare ?? configFile.trigger?.compare ?? DEFAULT_SKILLTEST_CONFIG.trigger.compare
+      )
     },
     eval: {
       numRuns: cliFlags.eval?.numRuns ?? configFile.eval?.numRuns ?? DEFAULT_SKILLTEST_CONFIG.eval.numRuns,
@@ -3145,6 +3400,18 @@ function extractCliConfigOverrides(command) {
       numQueries: getTypedOptionValue(command, "numQueries")
     };
   }
+  if ((command.name() === "trigger" || command.name() === "check") && command.getOptionValueSource("compare") === "cli") {
+    overrides.trigger = {
+      ...overrides.trigger,
+      compare: getTypedOptionValue(command, "compare")
+    };
+  }
+  if ((command.name() === "lint" || command.name() === "check") && command.getOptionValueSource("plugin") === "cli") {
+    overrides.lint = {
+      ...overrides.lint,
+      plugins: getTypedOptionValue(command, "plugin")
+    };
+  }
   if (command.name() === "check" && command.getOptionValueSource("minF1") === "cli") {
     overrides.trigger = {
       ...overrides.trigger,
@@ -3172,7 +3439,7 @@ async function resolveConfigContext(targetPath, cliFlags) {
       config: mergeConfigLayers(skillDirectoryConfig.configFile, cliFlags, skillDirectoryConfig.sourceDirectory)
     };
   }
-  const cwdConfigPath = path5.join(cwd, ".skilltestrc");
+  const cwdConfigPath = path6.join(cwd, ".skilltestrc");
   const cwdConfig = await loadConfigFromJsonFile(cwdConfigPath);
   if (cwdConfig) {
     return {
@@ -3393,6 +3660,7 @@ function createProvider(providerName, apiKeyOverride) {
 var triggerCliSchema = z8.object({
   queries: z8.string().optional(),
   saveQueries: z8.string().optional(),
+  compare: z8.array(z8.string().min(1)).optional(),
   seed: z8.number().int().optional(),
   concurrency: z8.number().int().min(1).optional(),
   html: z8.string().optional(),
@@ -3441,6 +3709,7 @@ async function handleTriggerCommand(targetPath, options) {
       provider,
       queries,
       numQueries: options.numQueries,
+      compare: options.compare,
       seed: options.seed,
       concurrency: options.concurrency,
       verbose: options.verbose
@@ -3459,7 +3728,7 @@ async function handleTriggerCommand(targetPath, options) {
         ...result,
         target: targetPath
       };
-      await fs8.writeFile(options.html, renderTriggerHtml(htmlResult), "utf8");
+      await fs9.writeFile(options.html, renderTriggerHtml(htmlResult), "utf8");
     }
   } catch (error) {
     spinner?.stop();
@@ -3468,7 +3737,7 @@ async function handleTriggerCommand(targetPath, options) {
   }
 }
 function registerTriggerCommand(program) {
-  program.command("trigger").description("Evaluate whether a skill description triggers correctly.").argument("<path-to-skill>", "Path to SKILL.md or skill directory").option("--model <model>", "Model to use").option("--provider <provider>", "LLM provider: anthropic|openai").option("--queries <path>", "Path to custom test queries JSON").option("--num-queries <n>", "Number of auto-generated queries", (value) => Number.parseInt(value, 10)).option("--seed <number>", "RNG seed for reproducible results", (value) => Number.parseInt(value, 10)).option("--concurrency <n>", "Maximum in-flight trigger requests", (value) => Number.parseInt(value, 10)).option("--html <path>", "Write an HTML report to the given file path").option("--save-queries <path>", "Save generated queries to a JSON file").option("--api-key <key>", "API key override").option("--verbose", "Show full model decisions").action(async (targetPath, _commandOptions, command) => {
+  program.command("trigger").description("Evaluate whether a skill description triggers correctly.").argument("<path-to-skill>", "Path to SKILL.md or skill directory").option("--model <model>", "Model to use").option("--provider <provider>", "LLM provider: anthropic|openai").option("--queries <path>", "Path to custom test queries JSON").option("--compare <path...>", "Path(s) to sibling skill directories to include as competitors").option("--num-queries <n>", "Number of auto-generated queries", (value) => Number.parseInt(value, 10)).option("--seed <number>", "RNG seed for reproducible results", (value) => Number.parseInt(value, 10)).option("--concurrency <n>", "Maximum in-flight trigger requests", (value) => Number.parseInt(value, 10)).option("--html <path>", "Write an HTML report to the given file path").option("--save-queries <path>", "Save generated queries to a JSON file").option("--api-key <key>", "API key override").option("--verbose", "Show full model decisions").action(async (targetPath, _commandOptions, command) => {
     const globalOptions = getGlobalCliOptions(command);
     const config = getResolvedConfig(command);
     const parsedCli = triggerCliSchema.safeParse(command.opts());
@@ -3483,6 +3752,7 @@ function registerTriggerCommand(program) {
       provider: config.provider,
       queries: parsedCli.data.queries,
       numQueries: config.trigger.numQueries,
+      compare: config.trigger.compare,
       saveQueries: parsedCli.data.saveQueries,
       seed: parsedCli.data.seed ?? config.trigger.seed,
       concurrency: config.concurrency,
@@ -3494,7 +3764,7 @@ function registerTriggerCommand(program) {
 }
 
 // src/commands/eval.ts
-import fs9 from "node:fs/promises";
+import fs10 from "node:fs/promises";
 import ora2 from "ora";
 import { z as z9 } from "zod";
 var evalCliSchema = z9.object({
@@ -3561,7 +3831,7 @@ async function handleEvalCommand(targetPath, options, command) {
         ...result,
         target: targetPath
       };
-      await fs9.writeFile(options.html, renderEvalHtml(htmlResult), "utf8");
+      await fs10.writeFile(options.html, renderEvalHtml(htmlResult), "utf8");
     }
   } catch (error) {
     spinner?.stop();
@@ -3600,7 +3870,7 @@ function registerEvalCommand(program) {
 }
 
 // src/commands/check.ts
-import fs10 from "node:fs/promises";
+import fs11 from "node:fs/promises";
 import ora3 from "ora";
 import { z as z10 } from "zod";
 
@@ -3613,7 +3883,7 @@ function calculateEvalAssertPassRate(result) {
 }
 async function runCheck(inputPath, options) {
   options.onStage?.("lint");
-  const lint = await runLinter(inputPath, { suppress: options.lintSuppress });
+  const lint = await runLinter(inputPath, { suppress: options.lintSuppress, plugins: options.lintPlugins });
   const lintPassed = !lintFails(lint, options.lintFailOn);
   let trigger = null;
   let evalResult = null;
@@ -3637,6 +3907,7 @@ async function runCheck(inputPath, options) {
         provider: options.provider,
         model: options.model,
         queries: options.queries,
+        compare: options.compare,
         numQueries: options.numQueries,
         seed: options.triggerSeed,
         concurrency: options.concurrency,
@@ -3698,8 +3969,10 @@ var checkCliSchema = z10.object({
   graderModel: z10.string().optional(),
   apiKey: z10.string().optional(),
   queries: z10.string().optional(),
+  compare: z10.array(z10.string().min(1)).optional(),
   seed: z10.number().int().optional(),
   prompts: z10.string().optional(),
+  plugin: z10.array(z10.string().min(1)).optional(),
   concurrency: z10.number().int().min(1).optional(),
   html: z10.string().optional(),
   saveResults: z10.string().optional(),
@@ -3708,6 +3981,9 @@ var checkCliSchema = z10.object({
 });
 var DEFAULT_ANTHROPIC_MODEL3 = "claude-sonnet-4-5-20250929";
 var DEFAULT_OPENAI_MODEL3 = "gpt-4.1-mini";
+function collectPluginPaths2(value, previous = []) {
+  return [...previous, value];
+}
 function resolveModel3(provider, model) {
   if (provider === "openai" && model === DEFAULT_ANTHROPIC_MODEL3) {
     return DEFAULT_OPENAI_MODEL3;
@@ -3758,7 +4034,9 @@ async function handleCheckCommand(targetPath, options, command) {
       graderModel,
       lintFailOn: options.lintFailOn,
       lintSuppress: options.lintSuppress,
+      lintPlugins: options.lintPlugins,
       queries,
+      compare: options.compare,
       numQueries: options.numQueries,
       triggerSeed: options.triggerSeed,
       prompts,
@@ -3794,7 +4072,7 @@ async function handleCheckCommand(targetPath, options, command) {
       );
     }
     if (options.html) {
-      await fs10.writeFile(options.html, renderCheckHtml(result), "utf8");
+      await fs11.writeFile(options.html, renderCheckHtml(result), "utf8");
     }
     process.exitCode = result.gates.overallPassed ? 0 : 1;
   } catch (error) {
@@ -3804,7 +4082,7 @@ async function handleCheckCommand(targetPath, options, command) {
   }
 }
 function registerCheckCommand(program) {
-  program.command("check").description("Run lint + trigger + eval with threshold-based quality gates.").argument("<path-to-skill>", "Path to SKILL.md or skill directory").option("--provider <provider>", "LLM provider: anthropic|openai").option("--model <model>", "Model for trigger/eval runs").option("--grader-model <model>", "Model used for grading (defaults to --model)").option("--api-key <key>", "API key override").option("--queries <path>", "Path to custom trigger queries JSON").option("--num-queries <n>", "Number of auto-generated trigger queries", (value) => Number.parseInt(value, 10)).option("--seed <number>", "RNG seed for reproducible results", (value) => Number.parseInt(value, 10)).option("--prompts <path>", "Path to eval prompts JSON").option("--concurrency <n>", "Maximum in-flight trigger/eval tasks", (value) => Number.parseInt(value, 10)).option("--html <path>", "Write an HTML report to the given file path").option("--min-f1 <n>", "Minimum required trigger F1 score (0-1)", (value) => Number.parseFloat(value)).option("--min-assert-pass-rate <n>", "Minimum required eval assertion pass rate (0-1)", (value) => Number.parseFloat(value)).option("--save-results <path>", "Save combined check results to JSON").option("--continue-on-lint-fail", "Continue trigger/eval stages even when lint has failures").option("--verbose", "Show detailed trigger/eval output sections").action(async (targetPath, _commandOptions, command) => {
+  program.command("check").description("Run lint + trigger + eval with threshold-based quality gates.").argument("<path-to-skill>", "Path to SKILL.md or skill directory").option("--provider <provider>", "LLM provider: anthropic|openai").option("--model <model>", "Model for trigger/eval runs").option("--grader-model <model>", "Model used for grading (defaults to --model)").option("--api-key <key>", "API key override").option("--queries <path>", "Path to custom trigger queries JSON").option("--compare <path...>", "Path(s) to sibling skill directories to include as competitors").option("--num-queries <n>", "Number of auto-generated trigger queries", (value) => Number.parseInt(value, 10)).option("--seed <number>", "RNG seed for reproducible results", (value) => Number.parseInt(value, 10)).option("--prompts <path>", "Path to eval prompts JSON").option("--plugin <path>", "Load a custom lint plugin file", collectPluginPaths2, []).option("--concurrency <n>", "Maximum in-flight trigger/eval tasks", (value) => Number.parseInt(value, 10)).option("--html <path>", "Write an HTML report to the given file path").option("--min-f1 <n>", "Minimum required trigger F1 score (0-1)", (value) => Number.parseFloat(value)).option("--min-assert-pass-rate <n>", "Minimum required eval assertion pass rate (0-1)", (value) => Number.parseFloat(value)).option("--save-results <path>", "Save combined check results to JSON").option("--continue-on-lint-fail", "Continue trigger/eval stages even when lint has failures").option("--verbose", "Show detailed trigger/eval output sections").action(async (targetPath, _commandOptions, command) => {
     const globalOptions = getGlobalCliOptions(command);
     const config = getResolvedConfig(command);
     const parsedCli = checkCliSchema.safeParse(command.opts());
@@ -3822,6 +4100,7 @@ function registerCheckCommand(program) {
         graderModel: parsedCli.data.graderModel,
         apiKey: parsedCli.data.apiKey,
         queries: parsedCli.data.queries,
+        compare: config.trigger.compare,
         numQueries: config.trigger.numQueries,
         prompts: parsedCli.data.prompts,
         minF1: config.trigger.threshold,
@@ -3831,6 +4110,7 @@ function registerCheckCommand(program) {
         html: parsedCli.data.html,
         lintFailOn: config.lint.failOn,
         lintSuppress: config.lint.suppress,
+        lintPlugins: config.lint.plugins,
         triggerSeed: parsedCli.data.seed ?? config.trigger.seed,
         saveResults: parsedCli.data.saveResults,
         continueOnLintFail: Boolean(parsedCli.data.continueOnLintFail),
@@ -3845,8 +4125,8 @@ function registerCheckCommand(program) {
 function resolveVersion() {
   try {
     const currentFilePath = fileURLToPath(import.meta.url);
-    const packageJsonPath = path6.resolve(path6.dirname(currentFilePath), "..", "package.json");
-    const raw = fs11.readFileSync(packageJsonPath, "utf8");
+    const packageJsonPath = path7.resolve(path7.dirname(currentFilePath), "..", "package.json");
+    const raw = fs12.readFileSync(packageJsonPath, "utf8");
     const parsed = JSON.parse(raw);
     return parsed.version ?? "0.0.0";
   } catch {