[zzxwill]

Standalone jury reply sent a prompt and wrote it to a Markdown artifact, but omitted its text from the reply.sent event. Saved conversations therefore rendered an empty judge message before the reviewer answer. Persist the sent text using the same event schema as autonomous replies.

The CLI regression exercises two local fake reviewers and verifies each saved prompt, reviewer isolation, judge attribution and answer. It failed on unchanged master, then passed after the fix. README, changelog and the command coverage map describe the behavior.

Validation:

• Focused reply tests: 6 passed.
• Full suite: 287 passed.
• Fresh package checks: passed (55 packaged files and installed CLI regressions).
• Actual Jury: exit 0, NO NEW FINDINGS / ROUND CLEAN on cdff20bf61404d1eff3ebd69be275d89cf54b00f (71.3 seconds). The reviewer sandbox blocked temporary test files with EPERM, so its review was static; independent full/package checks passed as recorded above.

Actual Jury command (from the isolated worktree):

GIT_AUTHOR_NAME=zzxwill GIT_COMMITTER_NAME=zzxwill node bin/jury.js review --dir /Users/bytedance/Documents/Codex/2026-10-20/task/codejury-post-141-audit --trunk master --reviewer codex --judge codex --rounds 2 --summary 'Review the focused standalone reply conversation-text fix. The CLI now saves the sent prompt in reply.sent, matching the autonomous loop. A real CLI regression failed on unchanged master and passes after the fix, checking two isolated reviewer threads and judge attribution. All 287 tests and fresh-package checks passed. Check correctness and scope; do not change unrelated behavior.' --web=false --push=false > ../reply-conversation-review-evidence/jury-final.log 2>&1

This fix applies to newly sent replies; it does not migrate older event logs.