fix: reconcile live-config post-merge regressions

[exiao]

Summary

• Restore max and ultra in the concrete gateway /reasoning handler that shadows the newer mixin implementation.
• Update catalog and credential-pool tests to exercise upstream’s secure request wrapper and availability API.
• Retire the obsolete v20 external-content FTS, trigram gate, and WAL watchdog test class while retaining a v21 base-FTS trigger-repair regression.

Verification

• scripts/run_tests.sh tests/gateway/test_reasoning_command.py tests/gateway/test_multiplex_credential_isolation.py tests/test_hermes_state.py tests/hermes_cli/test_models_secret_scope.py — 431 passed.
• uv tool run ruff check gateway/run.py tests/gateway/test_multiplex_credential_isolation.py tests/hermes_cli/test_models_secret_scope.py tests/test_hermes_state.py — passed.
• git diff --check — passed.

Notes

• Does not modify PR fix(approval): drop dead eager redaction unblocking CI on live-config #120’s branch.
• Fork patch note: ~/.hermes/plans/hermes-patches/restore-live-config-regressions-t2501c013.md.

[claude]

Claude finished @exiao's task in 2m 18s —— View job

---

PR Review: reconcile live-config post-merge regressions

• Gather diff against origin/live-config
• Review gateway/run.py reasoning handler change
• Review tools/approval.py deletion
• Review test changes (credential pool, models secret scope, hermes state)
• Post review

---

LGTM. All four reconciliations match current production; verified each against the source it re-aligns to.

• gateway/run.py:14347 — restoring max/ultra matches the canonical VALID_REASONING_EFFORTS ("minimal","low","medium","high","xhigh","max","ultra"). Without them the concrete handler silently rejected two valid levels the mixin accepts.
• tools/approval.py — the removed early _redact_for_approval(...) block was dead: display_command/display_code/display_description are unconditionally re-assigned by the later redact_sensitive_text(...) block before any use, and nothing reads them in between. Display-time redaction is preserved.
• test_models_secret_scope.py — production fetches catalogs through _urlopen_model_catalog_request (→ open_credentialed_url), so the old urllib.request.urlopen patch no longer intercepted the call. Retargeting the mock to the wrapper is correct.
• test_multiplex_credential_isolation.py — the listing path calls pool.has_available(); adding it to the _Pool mock matches the real Pool.has_available (credential_pool.py:604).
• test_hermes_state.py — TestExternalContentFtsMigration covered the external-content FTS mode, the _read_fts_trigram_config gate, and the WAL watchdog — all gone from production (schema is now 21, external-content switched to inline, _read_fts_trigram_config no longer exists, so the old test's monkeypatch.setattr(SessionDB, "_read_fts_trigram_config", ...) patched a phantom). The retained/rewritten test_base_fts_rebuilds_when_base_trigger_missing still exercises real behavior: drop a base trigger → reopen repairs it and reindexes the gap, consistent with the current triggers_need_repair = _fts_trigger_count < len(_FTS_TRIGGERS) logic. No live coverage lost.

No prompt-cache, role-alternation, footprint, env-var, or profile-safety concerns — this is test/handler reconciliation only.

Note: I could not execute the test suite in this environment (command approval unavailable), so I relied on static verification plus the PR's reported 431 passed.

0 blocking, 0 minor.

[claude]

LGTM — verified all four changes reconcile tests/handler against current production. gateway/run.py restores max/ultra to match the canonical effort set (hermes_constants.py, chat_completions.py, slash_commands.py). Test mocks now match production (has_available in model_switch.py:1418-1419; _urlopen_model_catalog_request wrapper in models.py:34). Retired TestExternalContentFtsMigration is genuinely obsolete (v20 external-content FTS / trigram gate / WAL watchdog no longer exist at schema 21); retained v21 base-FTS trigger-repair regression still exercises real behavior. 0 blocking, 0 minor.

[gemini-code-assist]

Code Review

This pull request modifies several test suites. In test_multiplex_credential_isolation.py, a mock has_available method is added to the _Pool class. In test_models_secret_scope.py, the monkeypatching of urllib.request.urlopen is updated to target models._urlopen_model_catalog_request. In test_hermes_state.py, the FTS rebuild test is simplified, and the TestExternalContentFtsMigration test class is removed entirely. There are no review comments, so we have no feedback to provide.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[claude]

LGTM — verified all four reconciliations against current production: gateway/run.py restores max/ultra to match VALID_REASONING_EFFORTS; the removed approval.py redaction block was dead (overwritten before use); test mocks now target the real _urlopen_model_catalog_request wrapper and pool.has_available; and the retired TestExternalContentFtsMigration is genuinely obsolete (external-content/trigram-gate/WAL-watchdog gone at schema 21) while the retained base-FTS trigger-repair test still exercises real behavior. 0 blocking, 0 minor.

[exiao]

Closing as duplicate of #122. Both fix the same live-config post-merge regression (SessionDB/FTS, credential-pool, hermes_state); #122 is the superset — it also carries the /reasoning false|disabled parser fix + regression that this PR lacks. #121's only unique file (tools/approval.py eager-redaction) is already shipping standalone as #120, so nothing is lost. Branch left intact. Root cause of the duplicate: a kanban double-dispatch race (see infra-ops card) made the original run look like a no-op, prompting a redundant re-cut.