@skill-harness/cli 0.32.2 → 0.32.4

package/dist/cli.js

@@ -6,6 +6,7 @@ import { discover, resolveSkill, loadSpec, parseSpec, appendScenario, parseModel
 import { getAdapter } from "@skill-harness/adapters";
 import { serveReview } from "./serve.js";
 import { cmdDecision } from "./decision.js";
+import { formatDiagnosis } from "./diagnose-format.js";
 const DEFAULT_MODEL = "fireworks:accounts/fireworks/models/deepseek-v4-pro";
 // The judge default lives in core (`defaultJudge()`), which resolves
 // SKILL_HARNESS_JUDGE over a baked value — it was duplicated in three places
@@ -768,7 +769,7 @@ export function help() {
   preflight <skill|all> --skills <root> [--specs <root>] eval checks (${free("preflight")})
   calibrate <manifest.json> [--run --judge p:m --out NEW_DIR] validate or judge reviewed controls
                      [--plan --judge p:m] [--only c1,c2] [--reuse PRIOR_DIR ...] exact historical reuse, no new subject calls
-  diagnose <run-dir> --skills <root> [--specs <root>] inspect saved evidence (${free("diagnose")})
+  diagnose <run-dir> --skills <root> [--specs <root>] [--format json|text] inspect saved evidence (${free("diagnose")})
   Paid commands accept --max-memory-mb N / --max-write-mb N for sampled Linux process limits.
   run/grade/calibrate save bounded private resources-*.json; unsupported host metrics stay unknown.
   In Pi: /skill-harness create <idea>             guided skill + optional agent + eval creation
@@ -807,12 +808,16 @@ export async function cmdPreflight(args) {
 }
 export async function cmdDiagnose(args) {
     if (!args._[0])
-        throw new Error("usage: skill-harness diagnose <run-dir> --skills <root> [--specs <root>]");
+        throw new Error("usage: skill-harness diagnose <run-dir> --skills <root> [--specs <root>] [--format json|text]");
+    const format = flagStr(args, "format") ?? "json";
+    if (format !== "json" && format !== "text")
+        throw new Error("diagnose --format must be json or text");
     const runDir = resolve(args._[0]);
     const recorded = readResults(runDir);
     const { skills, specs } = roots(args);
     const selected = resolveSkill(skills, recorded.skill, specs);
-    console.log(JSON.stringify(diagnoseRun(runDir, selected.dir, selected.specPath), null, 2));
+    const report = diagnoseRun(runDir, selected.dir, selected.specPath);
+    console.log(format === "text" ? formatDiagnosis(report) : JSON.stringify(report, null, 2));
 }
 export async function cmdCalibrate(args, adapterOverride) {
     if (!args._[0])

package/dist/diagnose-format.d.ts

@@ -0,0 +1,3 @@
+import type { Diagnosis } from "@skill-harness/core";
+/** Human view of the same offline report; never a second scoring policy. */
+export declare function formatDiagnosis(report: Diagnosis): string;

package/dist/diagnose-format.js

@@ -0,0 +1,32 @@
+/** Human view of the same offline report; never a second scoring policy. */
+export function formatDiagnosis(report) {
+    const grouped = report.criterionDimensions;
+    const lines = [
+        `${grouped.skill} · ${grouped.model} · ${grouped.mode}`,
+        "Saved criterion votes — current author-declared grouping, not a new assessment.",
+        "Counts are criterion observations, not passing answers. Mixed criteria remain indivisible.",
+        "No model calls; saved results and acceptance rules unchanged.",
+        `Coverage: ${grouped.coverage.availableVotes}/${grouped.coverage.expectedVotes} known-size votes available; ${grouped.coverage.unknownScenarioSizes} scenario sizes unknown.`,
+        `Mapping: ${grouped.metadataSha256}`,
+    ];
+    for (const scenario of grouped.scenarios) {
+        lines.push("", `${scenario.id}: recorded ${scenario.original.judge_verdict}; override ${scenario.original.override ?? "none"}; objective ${scenario.original.objective?.status ?? "not recorded"}`);
+        for (const [dimension, counts] of Object.entries(scenario.counts)) {
+            if (counts.pass + counts.fail + counts.unavailable > 0)
+                lines.push(`  ${dimension}: ${counts.pass} PASS, ${counts.fail} FAIL, ${counts.unavailable} unavailable`);
+        }
+        const reasons = new Set([scenario.unavailableReason, ...scenario.repetitions.map(rep => rep.unavailableReason)].filter(Boolean));
+        for (const reason of reasons)
+            lines.push(`  Unavailable: ${reason}`);
+    }
+    if (report.issues.length) {
+        lines.push("", "Evidence and next checks:");
+        for (const issue of report.issues)
+            lines.push(`- ${issue.scenario ? issue.scenario + ": " : ""}${issue.category}: ${issue.evidence}\n  Next: ${issue.next}`);
+    }
+    lines.push("", "Reuse plan:");
+    for (const item of report.reusePlan.scenarios)
+        lines.push(`- ${item.id}: ${item.action} — ${item.note}`);
+    return lines.join("\n");
+}
+//# sourceMappingURL=diagnose-format.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@skill-harness/cli",
-    "version": "0.32.2",
+    "version": "0.32.4",
     "description": "skill-harness CLI — run, grade, review, and lint agent-skill scenarios on the pi harness",
     "type": "module",
     "license": "MIT",
@@ -44,7 +44,7 @@
         "prepack": "node ../../scripts/release-pack.mjs --guard"
     },
     "dependencies": {
-        "@skill-harness/core": "0.32.2",
-        "@skill-harness/adapters": "0.32.2"
+        "@skill-harness/core": "0.32.4",
+        "@skill-harness/adapters": "0.32.4"
     }
 }