feat(review): carry the round ledger in the posted review body

packages/cli/package.json

@@ -48,12 +48,12 @@
     "@qwen-code/audio-capture": "file:../audio-capture",
     "@qwen-code/channel-base": "file:../channels/base",
     "@qwen-code/channel-dingtalk": "file:../channels/dingtalk",
-    "@qwen-code/channel-wecom": "file:../channels/wecom",
-    "@qwen-code/channel-qqbot": "file:../channels/qqbot",
+    "@qwen-code/channel-feishu": "file:../channels/feishu",
     "@qwen-code/channel-github": "file:../channels/github",
     "@qwen-code/channel-gitlab": "file:../channels/gitlab",
-    "@qwen-code/channel-feishu": "file:../channels/feishu",
+    "@qwen-code/channel-qqbot": "file:../channels/qqbot",
     "@qwen-code/channel-telegram": "file:../channels/telegram",
+    "@qwen-code/channel-wecom": "file:../channels/wecom",
     "@qwen-code/channel-weixin": "file:../channels/weixin",
     "@qwen-code/qwen-code-core": "file:../core",
     "@qwen-code/sdk": "file:../sdk-typescript",
@@ -75,8 +75,8 @@
     "jsonc-parser": "^3.3.1",
     "lowlight": "^3.3.0",
     "p-limit": "^7.3.0",
-    "proper-lockfile": "^4.1.2",
     "prompts": "^2.4.2",
+    "proper-lockfile": "^4.1.2",
     "react": "^19.2.4",
     "read-package-up": "^11.0.0",
     "shell-quote": "^1.9.0",
@@ -88,8 +88,9 @@
     "undici": "^7.28.0",
     "wrap-ansi": "^10.0.0",
     "ws": "^8.18.0",
-    "yauzl": "^2.10.0",
+    "yaml": "^2.8.1",
     "yargs": "^17.7.2",
+    "yauzl": "^2.10.0",
     "zod": "^3.23.8"
   },
   "devDependencies": {

packages/cli/src/commands/review.test.ts

@@ -50,6 +50,8 @@ describe('reviewCommand', () => {
       'agent-prompt',
       'build-test',
       'base-tree',
+      'test-delta',
+      'extract-step',
       'script-lint',
       'resolve-anchors',
       'check-coverage',

packages/cli/src/commands/review.ts

@@ -23,6 +23,8 @@ import { checkCoverageCommand } from './review/check-coverage.js';
 import { agentPromptCommand } from './review/agent-prompt.js';
 import { buildTestCommand } from './review/build-test.js';
 import { baseTreeCommand } from './review/base-tree.js';
+import { testDeltaCommand } from './review/test-delta.js';
+import { extractStepCommand } from './review/extract-step.js';
 import { scriptLintCommand } from './review/script-lint.js';
 import { submitCommand } from './review/submit.js';
 import { testEfficacyCommand } from './review/test-efficacy.js';
@@ -47,6 +49,8 @@ export const reviewCommand: CommandModule = {
       .command(agentPromptCommand)
       .command(buildTestCommand)
       .command(baseTreeCommand)
+      .command(testDeltaCommand)
+      .command(extractStepCommand)
       .command(scriptLintCommand)
       .command(resolveAnchorsCommand)
       .command(checkCoverageCommand)
@@ -58,7 +62,7 @@ export const reviewCommand: CommandModule = {
       .command(cleanupCommand)
       .demandCommand(
         1,
-        'Specify a subcommand: run, parse-args, fetch-pr, capture-local, plan-diff, pr-context, comment-status, load-rules, agent-prompt, build-test, base-tree, script-lint, resolve-anchors, check-coverage, presubmit, test-efficacy, test-plan, compose-review, submit, or cleanup.',
+        'Specify a subcommand: run, parse-args, fetch-pr, capture-local, plan-diff, pr-context, comment-status, load-rules, agent-prompt, build-test, base-tree, test-delta, extract-step, script-lint, resolve-anchors, check-coverage, presubmit, test-efficacy, test-plan, compose-review, submit, or cleanup.',
       )
       .version(false),
   handler: () => {

packages/cli/src/commands/review/compose-review.test.ts

@@ -20,6 +20,7 @@ import { promptRecordDir, briefPath } from './lib/prompt-record.js';
 import { getGhHost, setGhHost } from './lib/gh.js';
 import {
   composeReview,
+  buildLedger,
   scriptLintGate,
   testPlanGate,
   composeReviewCommand,
@@ -3106,3 +3107,44 @@ describe('testPlanGate — Test Plan rulings, disclosed but never capping', () =
     expect(notes[5]).toBe('and 3 more');
   });
 });
+
+describe('buildLedger', () => {
+  it('numbers findings round-scoped, inline first then body Criticals', () => {
+    const l = buildLedger(
+      3,
+      [
+        {
+          path: 'src/a.ts',
+          line: 12,
+          body: '**[Critical]**: double free\ndetail',
+        },
+        { path: 'src/b.ts', line: 4, body: '**[Suggestion]** untested guard' },
+        { path: 'src/c.ts', body: 'no marker — not a finding' },
+      ],
+      ['`src/d.ts` unanchorable blocker'],
+    );
+    expect(l.round).toBe(3);
+    expect(l.findings).toEqual([
+      {
+        id: 'R3-1',
+        sev: 'C',
+        file: 'src/a.ts',
+        line: 12,
+        title: 'double free',
+      },
+      {
+        id: 'R3-2',
+        sev: 'S',
+        file: 'src/b.ts',
+        line: 4,
+        title: 'untested guard',
+      },
+      {
+        id: 'R3-3',
+        sev: 'C',
+        file: '(body)',
+        title: '`src/d.ts` unanchorable blocker',
+      },
+    ]);
+  });
+});

packages/cli/src/commands/review/compose-review.ts

@@ -40,6 +40,11 @@ import {
 } from './lib/roster.js';
 import { diffHashOf, type ScriptLintReport } from './script-lint.js';
 import type { TestPlanReport } from './test-plan.js';
+import {
+  serializeLedger,
+  type Ledger,
+  type LedgerFinding,
+} from './lib/ledger.js';
 import {
   CRITICAL_PREFIX,
   SUGGESTION_PREFIX,
@@ -1640,6 +1645,51 @@ export const composeReviewCommand: CommandModule = {
       ...parsed,
       ...countInlineFindings(drafted),
     });
+    // Embed the next round's machine ledger in the body about to be posted —
+    // the durable copy every future environment can read back via pr-context.
+    // The round number comes from the side file pr-context recovered from the
+    // PREVIOUS posted round (+1), never from the model; a first review is
+    // round 1. PR-mode only: a local review has no PR to carry the marker.
+    try {
+      const planRaw = parsed.planPath
+        ? (JSON.parse(readFileSync(parsed.planPath, 'utf8')) as {
+            prNumber?: unknown;
+          })
+        : null;
+      const pr = planRaw?.prNumber;
+      const isPr =
+        (typeof pr === 'number' && Number.isInteger(pr) && pr > 0) ||
+        (typeof pr === 'string' && /^\d+$/.test(pr));
+      if (isPr && parsed.planPath) {
+        let prevRound = 0;
+        try {
+          const prev = JSON.parse(
+            readFileSync(
+              join(
+                dirname(parsed.planPath),
+                `qwen-review-pr-${pr}-prev-ledger.json`,
+              ),
+              'utf8',
+            ),
+          ) as Ledger;
+          if (Number.isInteger(prev.round) && prev.round > 0) {
+            prevRound = prev.round;
+          }
+        } catch {
+          // No previous posted round recovered: this is round 1.
+        }
+        const ledger = buildLedger(
+          prevRound + 1,
+          drafted as Array<{ path?: unknown; line?: unknown; body?: unknown }>,
+          toStringList(parsed.bodyCriticals, 'bodyCriticals'),
+        );
+        result.body = `${result.body}\n\n${serializeLedger(ledger)}`;
+      }
+    } catch {
+      // The ledger is a carry-forward convenience, never worth failing the
+      // verdict over. A round without a marker costs the NEXT round its
+      // memory, nothing else.
+    }
     // The exact terminal verdict, persisted beside the fields it is computed
     // from. `event` + `cappedBy` alone cannot reconstruct it — a presubmit
     // downgrade also depends on `downgraded`/`downgradedFrom` — and Step 8's
@@ -1674,6 +1724,49 @@ export const composeReviewCommand: CommandModule = {
 };
 
 /** The terminal verdict, in the words Step 6 is told to print. */
+/**
+ * The next round's ledger: every finding this review is posting as its own —
+ * the drafted inline comments plus the body Criticals. Low-confidence findings
+ * never reach either input (they are terminal-only), so the ledger holds only
+ * claims the review stands behind, which is what the next round re-asserts.
+ */
+export function buildLedger(
+  round: number,
+  drafted: Array<{ path?: unknown; line?: unknown; body?: unknown }>,
+  bodyCriticals: string[],
+): Ledger {
+  const findings: LedgerFinding[] = [];
+  for (const c of drafted) {
+    const body = typeof c.body === 'string' ? c.body : '';
+    const sev = body.startsWith('**[Critical]**')
+      ? ('C' as const)
+      : body.startsWith('**[Suggestion]**')
+        ? ('S' as const)
+        : null;
+    if (!sev) continue;
+    const title = body
+      .replace(/^\*\*\[(Critical|Suggestion)\]\*\*:?\s*/, '')
+      .split('\n')[0]
+      .trim();
+    findings.push({
+      id: `R${round}-${findings.length + 1}`,
+      sev,
+      file: typeof c.path === 'string' ? c.path : '(unknown)',
+      ...(typeof c.line === 'number' ? { line: c.line } : {}),
+      title,
+    });
+  }
+  for (const b of bodyCriticals) {
+    findings.push({
+      id: `R${round}-${findings.length + 1}`,
+      sev: 'C',
+      file: '(body)',
+      title: b.split('\n')[0].trim(),
+    });
+  }
+  return { v: 1, round, findings };
+}
+
 export function verdictLine(r: ComposeReviewResult): string {
   const label: Record<ReviewEvent, string> = {
     APPROVE: 'Approve',