feat(review): capture-tui — rendering claims get pixels, not prose

docs/users/features/code-review.md

@@ -252,6 +252,8 @@ Confirmed findings are canonicalized into `.qwen/tmp/qwen-review-<target>-findin
 
 The command validates on write: a duplicate id, a finding with no failure scenario, an empty locations array, or an unknown severity is an error rather than a silently mangled entry.
 
+Where do the images come from? For terminal-rendering claims, `qwen review capture-tui` drives the code under review in a **private tmux server** (it cannot touch your own tmux sessions), captures the pane bytes as `.ans`, and renders a `.png` via `freeze` when installed — degrading explicitly (`png` → `ans-only` → refused) and recording which rung it reached, because a verifier must say whether its verdict stands on pixels, bytes, or prose.
+
 ## Evidence Images in PR Comments
 
 GitHub's API cannot attach images to review comments, so `/review` can host evidence images (TUI screenshots, rendered-output comparisons) in a repository you designate and embed them by URL:

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

@@ -48,6 +48,7 @@ describe('reviewCommand', () => {
       'comment-body',
       'fetch-pr',
       'capture-local',
+      'capture-tui',
       'plan-diff',
       'repo-context',
       'pr-context',

packages/cli/src/commands/review.ts

@@ -16,6 +16,7 @@ import { findingsCommand } from './review/findings.js';
 import { recoverFindingsCommand } from './review/recover-findings.js';
 import { fetchPrCommand } from './review/fetch-pr.js';
 import { captureLocalCommand } from './review/capture-local.js';
+import { captureTuiCommand } from './review/capture-tui.js';
 import { planDiffCommand } from './review/plan-diff.js';
 import { repoContextCommand } from './review/repo-context.js';
 import { prContextCommand } from './review/pr-context.js';
@@ -65,6 +66,7 @@ export const reviewCommand: CommandModule = {
       .command(commentBodyCommand)
       .command(fetchPrCommand)
       .command(captureLocalCommand)
+      .command(captureTuiCommand)
       .command(planDiffCommand)
       .command(repoContextCommand)
       .command(prContextCommand)
@@ -98,7 +100,7 @@ export const reviewCommand: CommandModule = {
       .command(cleanupCommand)
       .demandCommand(
         1,
-        'Specify a subcommand: run, parse-args, match-remote, meta, issue-context, fetch-diff, comment-body, fetch-pr, capture-local, plan-diff, repo-context, pr-context, comment-status, load-rules, agent-prompt, emit-workflow, build-test, base-tree, scratch-tree, test-delta, drive, ab-drive, mock-provider, extract-step, script-lint, dedup-candidates, revert-hunk, resolve-anchors, check-coverage, cost-ledger, presubmit, test-efficacy, test-plan, findings, recover-findings, publish-assets, compose-review, save-artifact, submit, or cleanup.',
+        'Specify a subcommand: run, parse-args, match-remote, meta, issue-context, fetch-diff, comment-body, fetch-pr, capture-local, capture-tui, plan-diff, repo-context, pr-context, comment-status, load-rules, agent-prompt, emit-workflow, build-test, base-tree, scratch-tree, test-delta, drive, ab-drive, mock-provider, extract-step, script-lint, dedup-candidates, revert-hunk, resolve-anchors, check-coverage, cost-ledger, presubmit, test-efficacy, test-plan, findings, recover-findings, publish-assets, compose-review, save-artifact, submit, or cleanup.',
       )
       .version(false),
   handler: () => {

packages/cli/src/commands/review/agent-prompt.test.ts

@@ -4230,6 +4230,13 @@ describe('verify and reverse-audit briefs — the Step 4/5 methodology, in code'
     expect(p).toContain('go read the claimed source first');
   });
 
+  it('wires capture-tui into the verify brief — the producer its rendering claims cite', () => {
+    // The brief teaches terminal-rendering claims to run `review capture-tui`;
+    // if the block drops out of the prompt the command exists with no consumer.
+    const p = buildRoleBrief(PLAN, 'verify');
+    expect(p).toContain('"${QWEN_CODE_CLI:-qwen}" review capture-tui');
+  });
+
   it('the verify brief carries the #9789 do-not-refute list and the constructible rejection bar', () => {
     // The recall leak the finder-side RECALL rule closes has a verifier half:
     // "silence is better than noise" read as a confidence bar lets Step 4 drop