Query: # Seventh independent review: idempotent request rebuilds

The fourth terminal acceptance review rejected the candidate after all focused
tests passed. Continue the same self-update and correct this retry-path defect:

- `append_persistent_model_status()` is not idempotent when the same newest
  message is used to build another request (for example, a transient provider
  failure, empty-response retry, or request rebuild before a new transcript
  message). It computes `updated = base + "\n\n" + status` from wire content
  that already contains the previous projection, so every rebuild adds another
  `<agent_status>` block.
- A direct production-helper exercise repeated the call 20 times and produced
  20 status blocks, changing previously sent bytes and violating both the
  prompt-cache stability claim and the total output bound.
- Current tests always append a new transcript message before the next build
  and therefore miss same-message retries.

Make projection idempotent at the production persistence boundary. Preserve an
unmodified base sidecar or safely recognize/replace only a status projection
owned by this feature; do not delete user-authored text that merely resembles
status markup. Decide and document whether volatile status values remain fixed
for a retry of the same message or can be replaced, but previously sent prefix
bytes must not drift unexpectedly. Keep clean transcript content unchanged,
retain string-only fail-closed behavior, and durably persist the selected
sidecar by stable row identity.

Add a regression test that repeatedly builds requests with unchanged source
messages, including a persisted/reloaded case. Assert identical wire bytes and
exactly one owned status block. Retain realistic tool-call placement,
persistence-reload, adversarial-bound, and unsupported-type coverage.

Run the focused tests, existing sidecar/cache/turn-context regressions,
compilation, and `git diff --check`. Update `BOOK_SELF_EVOLUTION_REPORT.md` with
this seventh review and exact results. Do not edit the book, commit, push, or
claim downstream task improvement.

Initializing agent...
↻ Resumed session 20260802_031522_523cd2 (7 user messages, 191 total messages)
────────────────────────────────────────

  ┊ 💻 preparing terminal…
  ┊ 💻 $         python3 - <<'PY' from pathlib import Path for f in ['agent/model_status_context.py','tests/agent/test_model_status_context.py']: print('\n###',f) + 1 command  0.0s
  ┊ 💻 preparing terminal…
  ┊ 💻 $         python3 - <<'PY' from pathlib import Path p=Path('agent/model_status_context.py') + 3 commands  1.2s
  ┊ 💻 preparing terminal…
  ┊ 💻 preparing terminal…
  ┊ 💻 preparing terminal…
  ┊ 💻 $         uv run --with pytest pytest tests/agent/test_api_content_sidecar.py tests/run_agent/test_background_review_cache_parity.py tests/agent/test_turn_context.py -q  10.4s
  ┊ 💻 $         python3 -m py_compile agent/model_status_context.py agent/conversation_loop.py agent/agent_init.py run_agent.py hermes_state.py + 1 command  0.1s
  ┊ 💻 $         git status --short + 1 command  0.0s
  ┊ 💻 preparing terminal…
  ┊ 💻 $         python3 - <<'PY' from pathlib import Path p=Path('BOOK_SELF_EVOLUTION_REPORT.md') + 5 commands  0.1s

╭─ ⚕ Hermes ───────────────────────────────────────────────────────────────────╮
Corrected the seventh-review idempotency defect.

Changes:

- Status projections are now idempotent for the same message.
- The first non-empty string api_content sidecar is treated as the feature-owned projection and reused verbatim on retries, including after database reload.
- Later volatile status values are deliberately ignored for that message, preserving previously sent bytes.
- User-authored status-looking text in clean transcript content is never parsed or removed.
- Stable row-identity backfill remains supported through update_message_api_content().
- Added a regression test that:
  - rebuilds the same request 20 times;
  - checks identical wire bytes;
  - checks exactly one <agent_status> block;
  - closes and reopens SessionDB;
  - repeats the assertions after reload.
- Retained realistic tool-call placement, persistence reload, output bounds, and unsupported-type tests.

Verification:

- uv run --with pytest pytest tests/agent/test_model_status_context.py -q
  - Passed: 7 passed in 0.84s
- uv run --with pytest pytest tests/agent/test_api_content_sidecar.py tests/run_agent/test_background_review_cache_parity.py tests/agent/test_turn_context.py -q
  - Passed: 36 passed in 9.42s
- python3 -m py_compile agent/model_status_context.py agent/conversation_loop.py agent/agent_init.py run_agent.py hermes_state.py
  - Passed.
- git diff --check
  - Passed.

BOOK_SELF_EVOLUTION_REPORT.md includes the seventh review round and exact results. No book changes, commit, push, or downstream task-improvement claim was made.
╰──────────────────────────────────────────────────────────────────────────────╯

Resume this session with:
  hermes --resume 20260802_031522_523cd2

Session:        20260802_031522_523cd2
Duration:       42s
Messages:       203 (8 user, 187 tool calls)