vigiles 33.1.0 → 33.2.0

package/dist/test-coverage-files.d.ts

@@ -1,7 +1,7 @@
 import { type PluginLayout } from "./core/layout.js";
 import type { CoverageDecision, CoverageTier, Surface } from "./test-coverage.js";
 /**
- * The untested harness surfaces (skills / agents / hooks) in a file map — the
+ * The untested harness surfaces (skills / agents / hooks / output styles) in a file map — the
  * browser-safe twin of `findUntestedSurfaces`, returning the union `untested` list
  * plus the same per-tier split (`scanFiles` needs the counts). Same coverage rules
  * as the disk detector; the tier split is by suffix, exactly as on disk.
@@ -17,5 +17,7 @@ repoName: string): {
     decisions: readonly CoverageDecision[];
     harness: CoverageTier;
     evals: CoverageTier;
+    /** Surfaces opted out via `vigiles:ignore-test`; still part of the inventory. */
+    exemptSurfaces: readonly Surface[];
 };
 //# sourceMappingURL=test-coverage-files.d.ts.map

package/dist/test-coverage-files.js

@@ -34,6 +34,7 @@ exports.findUntestedSurfacesInFiles = findUntestedSurfacesInFiles;
 const posix_path_js_1 = require("./posix-path.js");
 const layout_js_1 = require("./core/layout.js");
 const coverage_evidence_js_1 = require("./coverage-evidence.js");
+const output_style_js_1 = require("./core/output-style.js");
 const source_kinds_js_1 = require("./source-kinds.js");
 const IGNORE_MARKER = "vigiles:ignore-test";
 /** Mirror of the DEFAULT_IGNORE globs (root-anchored), as a key predicate. */
@@ -136,6 +137,13 @@ function discoverAgents(files, layout) {
     }
     return out;
 }
+/** Twin of test-coverage.ts `discoverOutputStyles`: the harness decides, over the map. */
+function discoverOutputStyles(files, layout) {
+    const scan = (0, output_style_js_1.findOutputStyles)(layout, new Map(Object.entries(files).filter(([path]) => !isIgnored(path))));
+    return scan.kind === "not-supported"
+        ? []
+        : scan.styles.map((style) => (0, coverage_evidence_js_1.outputStyleSurface)(style.path, files[style.path] ?? ""));
+}
 /**
  * Hook-script paths referenced from a manifest's `hooks` block (file hooks only).
  *
@@ -226,8 +234,17 @@ function tierOf(considered, tests, globs) {
     }
     return { covered, untested, decisions };
 }
+/** Every surface kind in the map — twin of test-coverage.ts `discoverSurfaces`. */
+function discoverSurfaces(files, layout, repoName) {
+    return [
+        ...discoverSkills(files, layout, repoName),
+        ...discoverAgents(files, layout),
+        ...discoverHooks(files, layout),
+        ...discoverOutputStyles(files, layout),
+    ];
+}
 /**
- * The untested harness surfaces (skills / agents / hooks) in a file map — the
+ * The untested harness surfaces (skills / agents / hooks / output styles) in a file map — the
  * browser-safe twin of `findUntestedSurfaces`, returning the union `untested` list
  * plus the same per-tier split (`scanFiles` needs the counts). Same coverage rules
  * as the disk detector; the tier split is by suffix, exactly as on disk.
@@ -239,11 +256,7 @@ function findUntestedSurfacesInFiles(files, layout,
  * `name:`. Same value `scanFiles` already threads for the skill inventory.
  */
 repoName) {
-    const surfaces = [
-        ...discoverSkills(files, layout, repoName),
-        ...discoverAgents(files, layout),
-        ...discoverHooks(files, layout),
-    ];
+    const surfaces = discoverSurfaces(files, layout, repoName);
     const considered = surfaces.filter((s) => !s.ignored);
     const tests = discoverTests(files);
     // NO configured `{surface}` globs here, and that is a property of this twin
@@ -257,6 +270,7 @@ repoName) {
     return {
         untested: [...union.untested],
         decisions: union.decisions,
+        exemptSurfaces: surfaces.filter((s) => s.ignored),
         harness: tierOf(considered, tests.filter((t) => !(0, coverage_evidence_js_1.isEvalScript)((0, posix_path_js_1.basename)(t.path))), noConfiguredGlobs),
         evals: tierOf(considered, tests.filter((t) => (0, coverage_evidence_js_1.isEvalScript)((0, posix_path_js_1.basename)(t.path))), noConfiguredGlobs),
     };

package/dist/test-coverage.d.ts

@@ -64,7 +64,7 @@ import { type ExcludeSet } from "./exclude.js";
 import type { RepoPath } from "./core/frame.js";
 import { type PluginLayout } from "./core/layout.js";
 import { type CoverageEvidence, type EvidenceCounts } from "./coverage-evidence.js";
-export type SurfaceKind = "skill" | "agent" | "hook";
+export type SurfaceKind = "skill" | "agent" | "hook" | "output-style";
 export interface Surface {
     readonly kind: SurfaceKind;
     /** Repo-relative path to the surface file (SKILL.md / agent .md / hook script). */
@@ -119,6 +119,11 @@ export interface UntestedReport {
     readonly untested: readonly Surface[];
     /** Surfaces explicitly opted out via `vigiles:ignore-test`. */
     readonly exempt: number;
+    /**
+     * The exempt surfaces themselves. The marker waives a surface's test, not its
+     * existence, so an inventory counts them (an exempt style still ships).
+     */
+    readonly exemptSurfaces: readonly Surface[];
     /**
      * Test files still carrying the RETIRED `vigiles:covers` marker.
      *
@@ -201,6 +206,18 @@ export interface TestCoverageOptions {
     readonly agents?: boolean;
     /** Scan hook scripts referenced from plugin.json / settings.json. Default true. */
     readonly hooks?: boolean;
+    /**
+     * Scan output styles, in every folder the layout's `outputStyles` names.
+     * Default true; a layout with no output styles finds none either way.
+     */
+    readonly outputStyles?: boolean;
+    /**
+     * Every harness the repository declares, when it declares more than one.
+     * Surfaces are found in each of them, one per file: a repository can list a
+     * harness first and keep its skills, agents or styles in another's folder.
+     * Defaults to `[layout]`.
+     */
+    readonly harnessLayouts?: readonly PluginLayout[];
     /** Globs of test files that count as coverage. */
     readonly include?: readonly string[];
     /**

package/dist/test-coverage.js

@@ -78,6 +78,7 @@ const ts_runner_caps_js_1 = require("./ts-runner-caps.js");
 const glob_1 = require("glob");
 const glob_ignore_js_1 = require("./core/glob-ignore.js");
 const exclude_js_1 = require("./exclude.js");
+const output_style_js_1 = require("./core/output-style.js");
 const source_kinds_js_1 = require("./source-kinds.js");
 const layout_js_1 = require("./core/layout.js");
 const coverage_evidence_js_1 = require("./coverage-evidence.js");
@@ -249,6 +250,46 @@ function discoverAgents(basePath, ignore, layout) {
     }
     return out;
 }
+/**
+ * Output styles, by the layout's own rules: every file in a style home that
+ * the harness would load. {@link findOutputStyles} decides; this only reads the
+ * candidates off the disk, so it and the browser twin cannot disagree on which
+ * files are styles.
+ */
+function discoverOutputStyles(basePath, ignore, layout) {
+    const candidates = new Map((0, output_style_js_1.outputStyleHomes)(layout).flatMap((home) => (0, glob_1.globSync)(`${home}/**/*`, {
+        cwd: basePath,
+        ignore: [...ignore],
+        dot: true,
+        nodir: true,
+    }).map((path) => [path.split(node_path_1.sep).join("/"), ""])));
+    const scan = (0, output_style_js_1.findOutputStyles)(layout, candidates);
+    return scan.kind === "not-supported"
+        ? []
+        : scan.styles.map((style) => (0, coverage_evidence_js_1.outputStyleSurface)(style.path, read((0, node_path_1.join)(basePath, style.path))));
+}
+/**
+ * Every kind the options switch on, found under the bundle. Discovery gets only
+ * the frame-free ignore floor; the rule's and the repo's excludes are applied
+ * later, from `root`, by the caller.
+ */
+function discoverSurfaces(basePath, layout, options) {
+    const on = (flag) => flag !== false;
+    const found = (options.harnessLayouts ?? [layout]).flatMap((each) => [
+        ...(on(options.skills)
+            ? discoverSkills(basePath, [...DEFAULT_IGNORE], each)
+            : []),
+        ...(on(options.agents)
+            ? discoverAgents(basePath, [...DEFAULT_IGNORE], each)
+            : []),
+        ...(on(options.hooks) ? discoverHooks(basePath, each) : []),
+        ...(on(options.outputStyles)
+            ? discoverOutputStyles(basePath, DEFAULT_IGNORE, each)
+            : []),
+    ]);
+    // Two declared harnesses can read the same file: one surface per file.
+    return [...new Map(found.map((s) => [`${s.kind}\0${s.path}`, s])).values()];
+}
 /**
  * Hook-script paths referenced from a manifest's `hooks` block (file hooks only).
  *
@@ -419,6 +460,11 @@ function tierOf(considered, tests, index, tier, globs) {
 // ---------------------------------------------------------------------------
 // Public API
 // ---------------------------------------------------------------------------
+/** The exempt surfaces and their count: waived from testing, still in the inventory. */
+function exemptOf(surfaces) {
+    const exemptSurfaces = surfaces.filter((s) => s.ignored);
+    return { exempt: exemptSurfaces.length, exemptSurfaces };
+}
 /**
  * Find harness surfaces (skills / agents / hooks) that no test or eval covers.
  * A surface is covered by a colocated `*.{harness,eval}.mjs` OR any discovered
@@ -449,24 +495,15 @@ function findUntestedSurfaces(options) {
     // Discovery globs from the BUNDLE, so it gets only the frame-free floor; the
     // rule's patterns are root-relative and the repo's carry their own root, so
     // both are applied below, to the root-relative path.
-    const floor = [...DEFAULT_IGNORE];
-    const found = [];
-    if (options.skills !== false)
-        found.push(...discoverSkills(basePath, floor, layout));
-    if (options.agents !== false)
-        found.push(...discoverAgents(basePath, floor, layout));
-    if (options.hooks !== false)
-        found.push(...discoverHooks(basePath, layout));
+    const found = discoverSurfaces(basePath, layout, options);
     // Hooks were never subject to `ignore` (a compiled hook lives under the
     // default-ignored `.vigiles/`); keep that, and re-apply exclude to the kinds
     // whose discovery glob used it — now from `root`, where the patterns live.
     const surfaces = found
         .map(toRoot)
         .filter((s) => s.kind === "hook" || !excluded(s.path));
-    // Every skill/agent/hook is held to the requirement — only an explicit
-    // `vigiles:ignore-test` marker exempts a surface (a visible, deliberate skip).
+    // Only an explicit `vigiles:ignore-test` marker exempts a surface.
     const considered = surfaces.filter((s) => !s.ignored);
-    const exempt = surfaces.length - considered.length;
     const tests = discoverTests(root, globs, ignore);
     const split = partitionTests(tests);
     // The run record, if there is one. NO artifact ⇒ an empty index ⇒ every
@@ -487,7 +524,7 @@ function findUntestedSurfaces(options) {
         total: considered.length,
         covered: union.covered,
         untested: union.untested,
-        exempt,
+        ...exemptOf(surfaces),
         staleRuns: staleRunsFor(considered, runIndex),
         testExt: (0, test_file_ext_js_1.testFileExt)({
             configured: options.testExtension,
@@ -683,6 +720,19 @@ function evalTierQuestion(kind) {
                 `the tool contract, \`assertAgentOk\` for the typed result. Those two ` +
                 `calls drive Claude Code and take no \`evalDriver\`, so on another ` +
                 `harness this tier has no public dispatch yet.`);
+        case "output-style":
+            // A style is not selected by the model, so there is no firing to
+            // measure; the real-model question is whether a model handed the style
+            // FOLLOWS it. The deterministic test that covers it drove a scripted
+            // model, which proves delivery and nothing about following.
+            return (`A deterministic test covers it, but that test drove a SCRIPTED model: ` +
+                `it can show the style reached the model, never that a real model ` +
+                `follows it.\n` +
+                `See the \`test-harness\` skill: compare runs with and without the style ` +
+                `(\`measureArms\`), check the output with deterministic checks where the ` +
+                `rule has a checkable shape and \`judged\` where it does not, and gate ` +
+                `with \`assertRates\`. NOT \`measureTriggerRate\`: a style is not ` +
+                `selected by the model, so there is no trigger to measure.`);
         case "hook":
             // Nothing. A hook is deterministic by construction: `runHook` answers
             // "does it block event X?" at the unit tier, free, for every event type —
@@ -859,8 +909,8 @@ function formatUntestedReport(report) {
     // model budget.
     lines.push(`  Two gaps, two costs: ${String(report.harness.untested.length)} with no ` +
         `deterministic harness (free, every push) · ` +
-        `${String(report.evals.untested.length)} whose firing was never measured ` +
-        `(needs a real model, run on a schedule).`);
+        `${String(report.evals.untested.length)} never measured with a real model ` +
+        `(paid, run on a schedule).`);
     // What the surfaces that DID pass are resting on.
     if (provenance)
         lines.push(`  ${provenance}`);

package/dist/test.d.ts

@@ -62,7 +62,7 @@ export { loadHook } from "./load-hook.js";
 export { experimental_hookState } from "./hook-state-store.js";
 export type { HookStateHandle, SeedStateOptions } from "./hook-state-store.js";
 export type { StateFact, Duration } from "./core/hook-state.js";
-export { evalChecks, assertChecks, tool, toolWith, notTool, onlyTools, skill, output, hookFired, received, turns, wrote, didNotWrite, subagent, blocked, allowed, mcp, cost, latency, tokens, inputTokens, outputTokens, cacheTokens, } from "./check.js";
+export { evalChecks, assertChecks, tool, toolWith, notTool, onlyTools, skill, output, experimental_eachReply, experimental_replyCount, hookFired, received, turns, wrote, didNotWrite, subagent, blocked, allowed, mcp, cost, latency, tokens, inputTokens, outputTokens, cacheTokens, } from "./check.js";
 export type { ArgMatcher, Check, CheckJSON, CheckResult, JudgeFn, } from "./check.js";
 export { DISASTER_CATALOG, verifyGuardrail, unblockedDisasters, assertBlocksDisasters, formatGuardrailReport, experimental_alternateSpellings, } from "./guardrail-check.js";
 export type { DisasterEvent, DisasterCategory, GuardrailResult, VerifyGuardrailOptions, } from "./guardrail-check.js";
@@ -70,6 +70,8 @@ export { experimental_verifyPluginGuards, experimental_formatPluginGuardReport,
 export type { PluginGuardReport, SweptHook, SweptHookOutcome, NonCommandHookAction, VerifyPluginGuardsOptions, } from "./verify-plugin-guards.js";
 export * from "./tool-stub.js";
 export { runHarnessTest, runHarness, parseToolCalls, parseSubagents, parseResultEvent, parseOutput, parseHooks, decideSandbox, specTrusted, sandboxAvailable, } from "./harness-test.js";
+export { experimental_outputStyleArms } from "./output-style-arms.js";
+export type { OutputStyleArms } from "./output-style-arms.js";
 export type { HarnessTestSpec, Trace, SubagentTrace, HarnessTestResult, RunHarnessTestOptions, ModelTurn, ModelRequest, ToolCall, HookFire, HarnessTestDriver, SandboxMode, } from "./harness-test.js";
 export { commandsIn, mustInclude, mustNotInclude } from "./doc-commands.js";
 export type { DocCommand } from "./doc-commands.js";

package/dist/test.js

@@ -71,8 +71,8 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
     for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.parseSubagents = exports.parseToolCalls = exports.runHarness = exports.runHarnessTest = exports.experimental_formatPluginGuardReport = exports.experimental_verifyPluginGuards = exports.experimental_alternateSpellings = exports.formatGuardrailReport = exports.assertBlocksDisasters = exports.unblockedDisasters = exports.verifyGuardrail = exports.DISASTER_CATALOG = exports.cacheTokens = exports.outputTokens = exports.inputTokens = exports.tokens = exports.latency = exports.cost = exports.mcp = exports.allowed = exports.blocked = exports.subagent = exports.didNotWrite = exports.wrote = exports.turns = exports.received = exports.hookFired = exports.output = exports.skill = exports.onlyTools = exports.notTool = exports.toolWith = exports.tool = exports.assertChecks = exports.evalChecks = exports.experimental_hookState = exports.loadHook = exports.experimental_assertEmittedOk = exports.experimental_parseEmitted = exports.experimental_emitTool = exports.egressRoutes = exports.fileToolEvents = exports.propertyHook = exports.decideHook = exports.parseHookOutput = exports.runHook = exports.runScript = exports.cleanupTmpDir = exports.makeTmpDir = exports.recordCheck = void 0;
-exports.experimental_makeDockerRuntime = exports.experimental_dockerRuntime = exports.experimental_withServices = exports.experimental_startServices = exports.stubSkillBody = exports.parseClaudeRun = exports.formatTriggerRateReport = exports.formatEvalReport = exports.formatCheckReport = exports.checkReportToJUnit = exports.checkPromptDiversity = exports.assertPromptDiversity = exports.assertRates = exports.defineEval = exports.formatContainment = exports.compareContainment = exports.skillContract = exports.mustNotInclude = exports.mustInclude = exports.commandsIn = exports.sandboxAvailable = exports.specTrusted = exports.decideSandbox = exports.parseHooks = exports.parseOutput = exports.parseResultEvent = void 0;
+exports.runHarness = exports.runHarnessTest = exports.experimental_formatPluginGuardReport = exports.experimental_verifyPluginGuards = exports.experimental_alternateSpellings = exports.formatGuardrailReport = exports.assertBlocksDisasters = exports.unblockedDisasters = exports.verifyGuardrail = exports.DISASTER_CATALOG = exports.cacheTokens = exports.outputTokens = exports.inputTokens = exports.tokens = exports.latency = exports.cost = exports.mcp = exports.allowed = exports.blocked = exports.subagent = exports.didNotWrite = exports.wrote = exports.turns = exports.received = exports.hookFired = exports.experimental_replyCount = exports.experimental_eachReply = exports.output = exports.skill = exports.onlyTools = exports.notTool = exports.toolWith = exports.tool = exports.assertChecks = exports.evalChecks = exports.experimental_hookState = exports.loadHook = exports.experimental_assertEmittedOk = exports.experimental_parseEmitted = exports.experimental_emitTool = exports.egressRoutes = exports.fileToolEvents = exports.propertyHook = exports.decideHook = exports.parseHookOutput = exports.runHook = exports.runScript = exports.cleanupTmpDir = exports.makeTmpDir = exports.recordCheck = void 0;
+exports.experimental_makeDockerRuntime = exports.experimental_dockerRuntime = exports.experimental_withServices = exports.experimental_startServices = exports.stubSkillBody = exports.parseClaudeRun = exports.formatTriggerRateReport = exports.formatEvalReport = exports.formatCheckReport = exports.checkReportToJUnit = exports.checkPromptDiversity = exports.assertPromptDiversity = exports.assertRates = exports.defineEval = exports.formatContainment = exports.compareContainment = exports.skillContract = exports.mustNotInclude = exports.mustInclude = exports.commandsIn = exports.experimental_outputStyleArms = exports.sandboxAvailable = exports.specTrusted = exports.decideSandbox = exports.parseHooks = exports.parseOutput = exports.parseResultEvent = exports.parseSubagents = exports.parseToolCalls = void 0;
 // --- reporting: how much did this script actually do? ---
 // `vigiles test` can otherwise see only an exit code, so a file that runs NOTHING
 // prints the same `✓` as one that ran and passed (measured 2026-08-08 on a file
@@ -159,6 +159,8 @@ Object.defineProperty(exports, "notTool", { enumerable: true, get: function () {
 Object.defineProperty(exports, "onlyTools", { enumerable: true, get: function () { return check_js_1.onlyTools; } });
 Object.defineProperty(exports, "skill", { enumerable: true, get: function () { return check_js_1.skill; } });
 Object.defineProperty(exports, "output", { enumerable: true, get: function () { return check_js_1.output; } });
+Object.defineProperty(exports, "experimental_eachReply", { enumerable: true, get: function () { return check_js_1.experimental_eachReply; } });
+Object.defineProperty(exports, "experimental_replyCount", { enumerable: true, get: function () { return check_js_1.experimental_replyCount; } });
 Object.defineProperty(exports, "hookFired", { enumerable: true, get: function () { return check_js_1.hookFired; } });
 Object.defineProperty(exports, "received", { enumerable: true, get: function () { return check_js_1.received; } });
 Object.defineProperty(exports, "turns", { enumerable: true, get: function () { return check_js_1.turns; } });
@@ -213,6 +215,10 @@ Object.defineProperty(exports, "parseHooks", { enumerable: true, get: function (
 Object.defineProperty(exports, "decideSandbox", { enumerable: true, get: function () { return harness_test_js_1.decideSandbox; } });
 Object.defineProperty(exports, "specTrusted", { enumerable: true, get: function () { return harness_test_js_1.specTrusted; } });
 Object.defineProperty(exports, "sandboxAvailable", { enumerable: true, get: function () { return harness_test_js_1.sandboxAvailable; } });
+// The two arms of an output-style eval, after one free run proves the style
+// reaches the model. Free: the run uses the scripted model.
+var output_style_arms_js_1 = require("./output-style-arms.js");
+Object.defineProperty(exports, "experimental_outputStyleArms", { enumerable: true, get: function () { return output_style_arms_js_1.experimental_outputStyleArms; } });
 // A document's own rule about its own COMMANDS, made checkable. vigiles cannot
 // infer "always pass -g" from prose; what was missing was a cheap way to declare
 // it — measured at ~95 lines of markdown parsing around ~5 lines of rule, which

package/dist/vigilesrc.schema.json

@@ -728,6 +728,140 @@
             }
           ]
         },
+        "untested-output-style": {
+          "anyOf": [
+            {
+              "anyOf": [
+                {
+                  "type": "string",
+                  "const": "warn"
+                },
+                {
+                  "type": "string",
+                  "const": "error"
+                },
+                {
+                  "type": "boolean",
+                  "const": false
+                },
+                {
+                  "type": "string",
+                  "const": "off"
+                },
+                {
+                  "type": "number",
+                  "const": 0
+                },
+                {
+                  "type": "number",
+                  "const": 1
+                },
+                {
+                  "type": "number",
+                  "const": 2
+                },
+                {
+                  "type": "boolean",
+                  "const": true
+                }
+              ]
+            },
+            {
+              "type": "array",
+              "prefixItems": [
+                {
+                  "type": "string",
+                  "const": "warn"
+                },
+                {
+                  "type": "object",
+                  "properties": {
+                    "include": {
+                      "anyOf": [
+                        {
+                          "type": "string"
+                        },
+                        {
+                          "type": "array",
+                          "items": {
+                            "type": "string"
+                          }
+                        }
+                      ]
+                    },
+                    "exclude": {
+                      "anyOf": [
+                        {
+                          "type": "string"
+                        },
+                        {
+                          "type": "array",
+                          "items": {
+                            "type": "string"
+                          }
+                        }
+                      ]
+                    },
+                    "testExtension": {
+                      "type": "string"
+                    }
+                  },
+                  "additionalProperties": false
+                }
+              ],
+              "items": false,
+              "minItems": 2,
+              "maxItems": 2
+            },
+            {
+              "type": "array",
+              "prefixItems": [
+                {
+                  "type": "string",
+                  "const": "error"
+                },
+                {
+                  "type": "object",
+                  "properties": {
+                    "include": {
+                      "anyOf": [
+                        {
+                          "type": "string"
+                        },
+                        {
+                          "type": "array",
+                          "items": {
+                            "type": "string"
+                          }
+                        }
+                      ]
+                    },
+                    "exclude": {
+                      "anyOf": [
+                        {
+                          "type": "string"
+                        },
+                        {
+                          "type": "array",
+                          "items": {
+                            "type": "string"
+                          }
+                        }
+                      ]
+                    },
+                    "testExtension": {
+                      "type": "string"
+                    }
+                  },
+                  "additionalProperties": false
+                }
+              ],
+              "items": false,
+              "minItems": 2,
+              "maxItems": 2
+            }
+          ]
+        },
         "unmarked-refs": {
           "anyOf": [
             {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vigiles",
-  "version": "33.1.0",
+  "version": "33.2.0",
   "description": "Audit, test and measure the harness your AI agent runs on — grade your CLAUDE.md / AGENTS.md, skills, subagents and hooks, run them against a scripted model, and measure whether they actually fire.",
   "keywords": [
     "claude-code",