feat: eval report, diff, list commands

Implement the 3 read commands for eval framework:

- report: read eval-run from CAS, render formatted text
  (task, overall, config, judges table, thread ID)
- diff: side-by-side comparison with ▲/▼ delta indicators
  and config change markers
- list: scan @uwf/eval/*/latest variables, sort by timestamp desc,
  --task filter, --limit pagination

Architecture: pure formatting functions (format.ts) + data access
(read.ts) + thin CLI handlers. Types in types.ts.

11 new tests (formatReport, formatDiff, formatList, selectEntries)

Refs #72

packages/eval/__tests__/commands.test.ts

@@ -0,0 +1,171 @@
+import { bootstrap, createMemoryStore, putSchema } from "@ocas/core";
+import type { CasRef } from "@united-workforce/protocol";
+import { describe, expect, test } from "vitest";
+
+import {
+  formatDiff,
+  formatList,
+  formatReport,
+  readEvalEntries,
+  readEvalRun,
+  selectEntries,
+} from "../src/commands/index.js";
+import type { EvalRunPayload, EvalStore } from "../src/storage/index.js";
+import { EVAL_RUN_SCHEMA, setEvalLatest } from "../src/storage/index.js";
+
+function makeEvalStore(): EvalStore {
+  const store = createMemoryStore();
+  bootstrap(store);
+  return { store, varStore: store.var };
+}
+
+function makePayload(
+  task: string,
+  overall: number,
+  timestamp: number,
+  judges: EvalRunPayload["judges"] = [
+    {
+      name: "frontmatter-compliance",
+      score: 1.0,
+      weight: 0.6,
+      dataHash: "AAAAAAAAAAAAA" as CasRef,
+    },
+    { name: "token-stats", score: 0.5, weight: 0, dataHash: "BBBBBBBBBBBBB" as CasRef },
+  ],
+  config: EvalRunPayload["config"] = {
+    agent: "hermes",
+    model: "claude-sonnet-4",
+    engineVersion: "1.0.0",
+  },
+): EvalRunPayload {
+  return { task, config, threadId: "THREAD0123456789", judges, overall, timestamp };
+}
+
+/** Store an eval-run node in CAS and index it under @uwf/eval/<task>/latest. */
+function storeRun(evalStore: EvalStore, payload: EvalRunPayload): string {
+  const schemaHash = putSchema(evalStore.store, EVAL_RUN_SCHEMA);
+  const hash = evalStore.store.cas.put(schemaHash, payload);
+  setEvalLatest(evalStore.varStore, payload.task, hash);
+  return hash;
+}
+
+describe("formatReport", () => {
+  test("includes task, overall, config and judges", () => {
+    const payload = makePayload("fix-off-by-one", 0.8, Date.UTC(2026, 0, 2, 3, 4, 5));
+    const output = formatReport(payload, "RUNHASH123456");
+
+    expect(output).toContain("fix-off-by-one");
+    expect(output).toContain("0.8000");
+    expect(output).toContain("hermes");
+    expect(output).toContain("claude-sonnet-4");
+    expect(output).toContain("1.0.0");
+    expect(output).toContain("frontmatter-compliance");
+    expect(output).toContain("token-stats");
+    expect(output).toContain("THREAD0123456789");
+    expect(output).toContain("RUNHASH123456");
+  });
+
+  test("round-trips a stored run via readEvalRun", () => {
+    const evalStore = makeEvalStore();
+    const payload = makePayload("fix-off-by-one", 0.75, Date.now());
+    const hash = storeRun(evalStore, payload);
+
+    const loaded = readEvalRun(evalStore, hash);
+    expect(loaded).not.toBeNull();
+    const output = formatReport(loaded as EvalRunPayload, hash);
+    expect(output).toContain("fix-off-by-one");
+    expect(output).toContain("0.7500");
+  });
+
+  test("readEvalRun returns null for a missing hash", () => {
+    const evalStore = makeEvalStore();
+    expect(readEvalRun(evalStore, "NOPENOPENOPE0")).toBeNull();
+  });
+});
+
+describe("list", () => {
+  test("lists eval runs stored under different tasks", () => {
+    const evalStore = makeEvalStore();
+    storeRun(evalStore, makePayload("fix-off-by-one", 0.8, 2000));
+    storeRun(evalStore, makePayload("write-docs", 0.6, 1000));
+
+    const entries = readEvalEntries(evalStore);
+    expect(entries).toHaveLength(2);
+
+    const output = formatList(selectEntries(entries, null, 20));
+    expect(output).toContain("fix-off-by-one");
+    expect(output).toContain("write-docs");
+  });
+
+  test("sorts newest-first by timestamp", () => {
+    const evalStore = makeEvalStore();
+    storeRun(evalStore, makePayload("old-task", 0.5, 1000));
+    storeRun(evalStore, makePayload("new-task", 0.5, 2000));
+
+    const selected = selectEntries(readEvalEntries(evalStore), null, 20);
+    expect(selected[0]?.task).toBe("new-task");
+    expect(selected[1]?.task).toBe("old-task");
+  });
+
+  test("--task filter only shows the matching task", () => {
+    const evalStore = makeEvalStore();
+    storeRun(evalStore, makePayload("fix-off-by-one", 0.8, 2000));
+    storeRun(evalStore, makePayload("write-docs", 0.6, 1000));
+
+    const output = formatList(selectEntries(readEvalEntries(evalStore), "write-docs", 20));
+    expect(output).toContain("write-docs");
+    expect(output).not.toContain("fix-off-by-one");
+  });
+
+  test("--limit caps the number of rows", () => {
+    const evalStore = makeEvalStore();
+    storeRun(evalStore, makePayload("task-a", 0.8, 3000));
+    storeRun(evalStore, makePayload("task-b", 0.6, 2000));
+    storeRun(evalStore, makePayload("task-c", 0.4, 1000));
+
+    const selected = selectEntries(readEvalEntries(evalStore), null, 2);
+    expect(selected).toHaveLength(2);
+    expect(selected.map((e) => e.task)).toEqual(["task-a", "task-b"]);
+  });
+
+  test("empty store renders a placeholder", () => {
+    const evalStore = makeEvalStore();
+    const output = formatList(selectEntries(readEvalEntries(evalStore), null, 20));
+    expect(output).toContain("(no eval runs found)");
+  });
+});
+
+describe("formatDiff", () => {
+  test("shows an upward delta when B scores higher", () => {
+    const a = makePayload("fix-off-by-one", 0.6, 1000);
+    const b = makePayload("fix-off-by-one", 0.8, 2000);
+    const output = formatDiff(a, "HASHA00000000", b, "HASHB00000000");
+
+    expect(output).toContain("▲");
+    expect(output).toContain("HASHA00000000");
+    expect(output).toContain("HASHB00000000");
+  });
+
+  test("shows a downward delta when B scores lower", () => {
+    const a = makePayload("fix-off-by-one", 0.9, 1000);
+    const b = makePayload("fix-off-by-one", 0.4, 2000);
+    const output = formatDiff(a, "HASHA00000000", b, "HASHB00000000");
+    expect(output).toContain("▼");
+  });
+
+  test("marks differing config values", () => {
+    const a = makePayload("fix-off-by-one", 0.6, 1000, undefined, {
+      agent: "hermes",
+      model: "claude-sonnet-4",
+      engineVersion: "1.0.0",
+    });
+    const b = makePayload("fix-off-by-one", 0.6, 2000, undefined, {
+      agent: "claude-code",
+      model: "claude-sonnet-4",
+      engineVersion: "1.0.0",
+    });
+    const output = formatDiff(a, "HASHA00000000", b, "HASHB00000000");
+    expect(output).toContain("≠");
+    expect(output).toContain("claude-code");
+  });
+});

packages/eval/src/commands/diff.ts

@@ -1,11 +1,38 @@
+import { createLogger } from "@united-workforce/util";
 import type { Command } from "commander";
 
+import { createEvalStore } from "../storage/index.js";
+import { formatDiff } from "./format.js";
+import { readEvalRun } from "./read.js";
+
+const log = createLogger({ sink: { kind: "stderr" } });
+const LOG_DIFF = "D3WZ8N5T";
+
 export function registerDiffCommand(program: Command): void {
   program
     .command("diff <hash1> <hash2>")
     .description("Compare two eval runs side-by-side")
-    .action(async (_hash1: string, _hash2: string) => {
+    .action(async (hash1: string, hash2: string) => {
-      process.stderr.write("uwf-eval diff: not yet implemented\n");
+      try {
-      process.exitCode = 1;
+        const evalStore = await createEvalStore();
+        const payloadA = readEvalRun(evalStore, hash1);
+        if (payloadA === null) {
+          process.stderr.write(`eval run not found: ${hash1}\n`);
+          process.exitCode = 1;
+          return;
+        }
+        const payloadB = readEvalRun(evalStore, hash2);
+        if (payloadB === null) {
+          process.stderr.write(`eval run not found: ${hash2}\n`);
+          process.exitCode = 1;
+          return;
+        }
+        log(LOG_DIFF, `diff a=${hash1} b=${hash2}`);
+        process.stdout.write(formatDiff(payloadA, hash1, payloadB, hash2));
+      } catch (e) {
+        const message = e instanceof Error ? e.message : String(e);
+        process.stderr.write(`${message}\n`);
+        process.exitCode = 1;
+      }
     });
 }

packages/eval/src/commands/format.ts

@@ -0,0 +1,148 @@
+import type { EvalRunPayload } from "../storage/index.js";
+import type { EvalListEntry } from "./types.js";
+
+const NAME_WIDTH = 28;
+const SCORE_WIDTH = 10;
+const TIMESTAMP_WIDTH = 26;
+
+/** Format a 0..1 score (or weight) with fixed precision. */
+function formatScore(value: number): string {
+  return value.toFixed(4);
+}
+
+/** Human-readable ISO-8601 timestamp from epoch milliseconds. */
+function formatTimestamp(ms: number): string {
+  return new Date(ms).toISOString();
+}
+
+/** Right-pad to a fixed column width (with a trailing space if already full). */
+function pad(value: string, width: number): string {
+  return value.length >= width ? `${value} ` : value.padEnd(width);
+}
+
+/** Directional indicator for a score delta (B relative to A). */
+function formatDelta(delta: number): string {
+  if (delta > 0) {
+    return `▲ +${formatScore(delta)}`;
+  }
+  if (delta < 0) {
+    return `▼ ${formatScore(delta)}`;
+  }
+  return `= ${formatScore(0)}`;
+}
+
+/** Render a single eval run as a human-readable report. */
+export function formatReport(payload: EvalRunPayload, runHash: string): string {
+  const lines: string[] = [];
+  lines.push("=== Eval Report ===");
+  lines.push(`Task:       ${payload.task}`);
+  lines.push(`Overall:    ${formatScore(payload.overall)}`);
+  lines.push(`Timestamp:  ${formatTimestamp(payload.timestamp)}`);
+  lines.push("");
+  lines.push("Config:");
+  lines.push(`  Agent:    ${payload.config.agent}`);
+  lines.push(`  Model:    ${payload.config.model}`);
+  lines.push(`  Engine:   ${payload.config.engineVersion}`);
+  lines.push("");
+  lines.push("Judges:");
+  lines.push(`  ${pad("NAME", NAME_WIDTH)}${pad("SCORE", SCORE_WIDTH)}WEIGHT`);
+  for (const judge of payload.judges) {
+    lines.push(
+      `  ${pad(judge.name, NAME_WIDTH)}${pad(formatScore(judge.score), SCORE_WIDTH)}${formatScore(judge.weight)}`,
+    );
+  }
+  lines.push("");
+  lines.push(`Thread:     ${payload.threadId}`);
+  lines.push(`Run:        ${runHash}`);
+  return `${lines.join("\n")}\n`;
+}
+
+/** Render a side-by-side comparison of two eval runs. */
+export function formatDiff(
+  payloadA: EvalRunPayload,
+  hashA: string,
+  payloadB: EvalRunPayload,
+  hashB: string,
+): string {
+  const lines: string[] = [];
+  lines.push("=== Eval Diff ===");
+  lines.push(`A: ${hashA}  (${payloadA.task})`);
+  lines.push(`B: ${hashB}  (${payloadB.task})`);
+  lines.push("");
+
+  const overallDelta = payloadB.overall - payloadA.overall;
+  lines.push("Overall:");
+  lines.push(
+    `  A=${formatScore(payloadA.overall)}  B=${formatScore(payloadB.overall)}  ${formatDelta(overallDelta)}`,
+  );
+  lines.push("");
+
+  lines.push("Config:");
+  lines.push(configLine("Agent", payloadA.config.agent, payloadB.config.agent));
+  lines.push(configLine("Model", payloadA.config.model, payloadB.config.model));
+  lines.push(configLine("Engine", payloadA.config.engineVersion, payloadB.config.engineVersion));
+  lines.push("");
+
+  lines.push("Judges:");
+  lines.push(`  ${pad("NAME", NAME_WIDTH)}${pad("A", SCORE_WIDTH)}${pad("B", SCORE_WIDTH)}DELTA`);
+  const scoresA = new Map(payloadA.judges.map((judge) => [judge.name, judge.score]));
+  const scoresB = new Map(payloadB.judges.map((judge) => [judge.name, judge.score]));
+  for (const name of unionJudgeNames(payloadA, payloadB)) {
+    const scoreA = scoresA.get(name);
+    const scoreB = scoresB.get(name);
+    const cellA = scoreA === undefined ? "—" : formatScore(scoreA);
+    const cellB = scoreB === undefined ? "—" : formatScore(scoreB);
+    const delta = scoreA !== undefined && scoreB !== undefined ? formatDelta(scoreB - scoreA) : "";
+    lines.push(
+      `  ${pad(name, NAME_WIDTH)}${pad(cellA, SCORE_WIDTH)}${pad(cellB, SCORE_WIDTH)}${delta}`,
+    );
+  }
+  return `${lines.join("\n")}\n`;
+}
+
+/** Render a table of indexed eval runs. */
+export function formatList(entries: ReadonlyArray<EvalListEntry>): string {
+  const lines: string[] = [];
+  lines.push(
+    `  ${pad("TASK", NAME_WIDTH)}${pad("OVERALL", SCORE_WIDTH)}${pad("TIMESTAMP", TIMESTAMP_WIDTH)}HASH`,
+  );
+  if (entries.length === 0) {
+    lines.push("  (no eval runs found)");
+  }
+  for (const entry of entries) {
+    lines.push(
+      `  ${pad(entry.task, NAME_WIDTH)}${pad(formatScore(entry.overall), SCORE_WIDTH)}${pad(formatTimestamp(entry.timestamp), TIMESTAMP_WIDTH)}${entry.hash}`,
+    );
+  }
+  return `${lines.join("\n")}\n`;
+}
+
+/** Sort newest-first, then apply optional task filter and result limit. */
+export function selectEntries(
+  entries: ReadonlyArray<EvalListEntry>,
+  task: string | null,
+  limit: number | null,
+): EvalListEntry[] {
+  const sorted = [...entries].sort((a, b) => b.timestamp - a.timestamp);
+  const filtered = task !== null ? sorted.filter((entry) => entry.task === task) : sorted;
+  return limit !== null ? filtered.slice(0, limit) : filtered;
+}
+
+/** Ordered union of judge names: A's order first, then B-only names. */
+function unionJudgeNames(payloadA: EvalRunPayload, payloadB: EvalRunPayload): string[] {
+  const names: string[] = [];
+  const seen = new Set<string>();
+  for (const judge of [...payloadA.judges, ...payloadB.judges]) {
+    if (!seen.has(judge.name)) {
+      seen.add(judge.name);
+      names.push(judge.name);
+    }
+  }
+  return names;
+}
+
+/** One config row: `=` when equal, `≠` otherwise. */
+function configLine(label: string, valueA: string, valueB: string): string {
+  const marker = valueA === valueB ? "=" : "≠";
+  return `  ${pad(`${label}:`, SCORE_WIDTH)}${marker} A=${valueA}  B=${valueB}`;
+}

packages/eval/src/commands/index.ts

@@ -1,4 +1,7 @@
 export { registerDiffCommand } from "./diff.js";
+export { formatDiff, formatList, formatReport, selectEntries } from "./format.js";
 export { registerListCommand } from "./list.js";
+export { readEvalEntries, readEvalRun } from "./read.js";
 export { registerReportCommand } from "./report.js";
 export { registerRunCommand } from "./run.js";
+export type { EvalListEntry } from "./types.js";

packages/eval/src/commands/list.ts

@@ -1,13 +1,43 @@
+import { createLogger } from "@united-workforce/util";
 import type { Command } from "commander";
 
+import { createEvalStore } from "../storage/index.js";
+import { formatList, selectEntries } from "./format.js";
+import { readEvalEntries } from "./read.js";
+
+const log = createLogger({ sink: { kind: "stderr" } });
+const LOG_LIST = "L5KX9R2B";
+
+type ListCliOptions = {
+  task: string | undefined;
+  limit: string;
+};
+
 export function registerListCommand(program: Command): void {
   program
     .command("list")
     .description("List past eval runs")
     .option("--task <name>", "filter by task name")
     .option("--limit <n>", "max results", "20")
-    .action(async (_opts: Record<string, unknown>) => {
+    .action(async (opts: ListCliOptions) => {
-      process.stderr.write("uwf-eval list: not yet implemented\n");
+      const limit = Number.parseInt(opts.limit, 10);
-      process.exitCode = 1;
+      if (!Number.isInteger(limit) || limit < 1) {
+        process.stderr.write("--limit must be a positive integer\n");
+        process.exitCode = 1;
+        return;
+      }
+
+      try {
+        const evalStore = await createEvalStore();
+        const entries = readEvalEntries(evalStore);
+        const task = opts.task ?? null;
+        const selected = selectEntries(entries, task, limit);
+        log(LOG_LIST, `list task=${task ?? "*"} found=${entries.length} shown=${selected.length}`);
+        process.stdout.write(formatList(selected));
+      } catch (e) {
+        const message = e instanceof Error ? e.message : String(e);
+        process.stderr.write(`${message}\n`);
+        process.exitCode = 1;
+      }
     });
 }

packages/eval/src/commands/read.ts

@@ -0,0 +1,41 @@
+import type { EvalRunPayload, EvalStore } from "../storage/index.js";
+import type { EvalListEntry } from "./types.js";
+
+/** Variable prefix and suffix for eval run pointers (`@uwf/eval/<task>/latest`). */
+const EVAL_VAR_PREFIX = "@uwf/eval/";
+const EVAL_VAR_SUFFIX = "/latest";
+
+/** Read a single eval-run payload from CAS. Returns null when the node is absent. */
+export function readEvalRun(evalStore: EvalStore, hash: string): EvalRunPayload | null {
+  const node = evalStore.store.cas.get(hash);
+  if (node === null) {
+    return null;
+  }
+  return node.payload as EvalRunPayload;
+}
+
+/**
+ * Read every indexed eval run by scanning `@uwf/eval/*\/latest` variables and
+ * loading the referenced CAS node. Dangling pointers are skipped.
+ */
+export function readEvalEntries(evalStore: EvalStore): EvalListEntry[] {
+  const { store, varStore } = evalStore;
+  const entries: EvalListEntry[] = [];
+  for (const variable of varStore.list()) {
+    if (!variable.name.startsWith(EVAL_VAR_PREFIX) || !variable.name.endsWith(EVAL_VAR_SUFFIX)) {
+      continue;
+    }
+    const node = store.cas.get(variable.value);
+    if (node === null) {
+      continue;
+    }
+    const payload = node.payload as EvalRunPayload;
+    entries.push({
+      task: payload.task,
+      overall: payload.overall,
+      timestamp: payload.timestamp,
+      hash: variable.value,
+    });
+  }
+  return entries;
+}

packages/eval/src/commands/report.ts

@@ -1,11 +1,32 @@
+import { createLogger } from "@united-workforce/util";
 import type { Command } from "commander";
 
+import { createEvalStore } from "../storage/index.js";
+import { formatReport } from "./format.js";
+import { readEvalRun } from "./read.js";
+
+const log = createLogger({ sink: { kind: "stderr" } });
+const LOG_REPORT = "R7QP2M4K";
+
 export function registerReportCommand(program: Command): void {
   program
     .command("report <hash>")
     .description("Show eval run results")
-    .action(async (_hash: string) => {
+    .action(async (hash: string) => {
-      process.stderr.write("uwf-eval report: not yet implemented\n");
+      try {
-      process.exitCode = 1;
+        const evalStore = await createEvalStore();
+        const payload = readEvalRun(evalStore, hash);
+        if (payload === null) {
+          process.stderr.write(`eval run not found: ${hash}\n`);
+          process.exitCode = 1;
+          return;
+        }
+        log(LOG_REPORT, `report task=${payload.task} hash=${hash}`);
+        process.stdout.write(formatReport(payload, hash));
+      } catch (e) {
+        const message = e instanceof Error ? e.message : String(e);
+        process.stderr.write(`${message}\n`);
+        process.exitCode = 1;
+      }
     });
 }

packages/eval/src/commands/types.ts

@@ -0,0 +1,9 @@
+import type { CasRef } from "@united-workforce/protocol";
+
+/** Summary row for the `list` command: one indexed eval run. */
+export type EvalListEntry = {
+  task: string;
+  overall: number;
+  timestamp: number;
+  hash: CasRef;
+};