…146)

T1 of the #143 thin-client refactor scope. Changes ``HERMES_DEFER_PERSONA``
from a binary on/off toggle to a tri-state with provider-aware
default:

| Env value | Behavior |
|---|---|
| Truthy (1/true/yes/on) | Defer (same as pre-T1) |
| Falsy (0/false/no/off) | Opt-out — keep baked hermes persona (NEW) |
| Unset/empty | Provider-aware: devagentic-local defers; others don't (T1 default-flip) |
| Unknown value | Falls back to provider default (doctor warns on typo) |

## Rationale

Per #143: when devagentic owns the model role + R5 workflow-preamble,
hermes-side "Hermes Agent by Nous Research" identity is a competing
voice. Devagentic-emitted system prompts should be authoritative
for devagentic-local sessions. Non-devagentic-local sessions
(OpenRouter, Anthropic, etc.) keep the legacy persona — they
don't have an upstream preamble source to defer to.

## Implementation

- ``agent/system_prompt.py::_resolve_persona_deferred(agent=None)``
  now accepts optional agent, reads provider from
  ``getattr(agent, "provider", "")``.
- New constant ``_DEFER_PERSONA_FALSY = {"0", "false", "no", "off"}``
  enables explicit opt-out (previously fell through to default-false).
- Call site in ``build_system_prompt_parts`` threads ``agent`` to
  the resolver.
- Backward compat: ``_resolve_persona_deferred()`` (no args) still
  works — returns False on unset (no provider context).

## Doctor probe

Updated ``_check_persona_deferred_env`` to tri-state:
- Truthy → ``check_ok`` (defer note)
- Falsy → ``check_info`` (explicit opt-out, legacy behavior)
- Unknown → ``check_warn`` (typo; full truthy+falsy legend in detail)

## Tests

- 8 new T1 tests in
  ``tests/agent/test_system_prompt_persona_deferred.py``:
  default-flip on devagentic-local; other providers unchanged;
  explicit-falsy overrides default-flip (4 parametrized);
  explicit-truthy unchanged on other providers; unknown env falls
  back to provider default; legacy no-agent-arg unchanged; e2e
  SOUL.md drop on devagentic-local; e2e SOUL.md kept on
  openrouter.
- Updated doctor probe tests:
  ``test_falsy_emits_info_with_opt_out_note`` (4 parametrized
  falsy values) and ``test_unknown_value_emits_warn_with_full_legend``
  (3 parametrized unknown values) replace the prior
  ``test_non_truthy_emits_warn_with_valid_sample``.
- 125 total green across affected suites (persona + intent + doctor
  probes + T3 + diag-env-gate).

## Composition

Per #143 sequencing (T3 → T1 → T2-gated → T2-default-flip):
- T3 / PR #145 (clarify default-out) merged ✓
- **This PR: T1 (persona default-flip)**
- Next: T2 (env-gated empty-content recovery removal)
- Then: T2 default-flip
- Later: T4-T6

…MPTY_RECOVERY (refs #143) (#147)

T2 of the #143 thin-client refactor scope. Devagentic-side cascade
(NousResearch#324) + runaway detector (NousResearch#345-348) + exec-terminus (NousResearch#349-354)
now cover the empty-content recovery layer with full intent /
role / dispatch-trace context. Hermes-side recovery layered on top
caused 3×4 dispatch stacking (#118) + mode confusion + invisible
swallow points (#133 debug funnel).

This patch default-flips: the four hermes-side recovery paths are
short-circuited unless ``HERMES_LEGACY_EMPTY_RECOVERY`` is set to
a truthy value. Legacy users opt-in to keep the pre-T2 behavior.

## Gated paths (all skipped when env unset)

| Path | Source | Loc |
|---|---|---|
| ``_finish_wants_tools`` synthetic recovery | PR #108 / #99 | line ~3618 |
| ``_post_tool_empty_retried`` nudge | pre-existing NousResearch#9400-class | line ~3749 |
| ``_structural_empty`` synthetic recovery | PR #69 / #67 | line ~3880 |
| 3-retry empty-content loop | pre-#67 | line ~3927 |

When env unset (default), empty responses fall through to either
the fallback-chain provider switch (if configured) or the clean
``(empty)`` terminal with ``_empty_terminal_sentinel=True``.

## Preserved through T2 (verified by test_t2_legacy_empty_recovery_gate.py)

- **PR #119 cascade_exhausted short-circuit** — hermes correctly
  deferring to devagentic's sentinel; NOT recovery. Source-level
  test asserts the ``if _cascade_err:`` block is NOT prefixed by
  ``_legacy_recovery_on``.
- **PR #122/#125 raw tool_calls fallback** — pre-recovery wire
  parsing in transports/chat_completions.py. Source-level test
  asserts the helper name doesn't appear in conversation_loop
  (lives elsewhere; untouched).
- **PR #131/#136/#138/#141 diagnostics** — env-gated via
  ``HERMES_DIAG_RAW_CAPTURE`` (#140), independent of this env.

## Operator deploy

Default behavior changes: empty responses surface cleanly (no
synthetic re-prompt). To preserve pre-T2 behavior:

```bash
export HERMES_LEGACY_EMPTY_RECOVERY=1
```

The legacy escape hatch is intended as a temporary safety net
while operators validate the thin-client architecture. Once the
devagentic-side cascade is universally deployed + observed to
cover all empty-content cases, the legacy gate can be removed in
a follow-up (the env var stays as a no-op for backward compat).

## Tests

- 23 new tests in
  ``tests/agent/test_t2_legacy_empty_recovery_gate.py``:
  resolver default-false / empty-false / 6 truthy / 6 falsy-or-
  unknown; source-level gate assertions for each of the four
  recovery branches; preserved-path assertions for cascade_exhausted
  + raw tool_calls fallback location; default-off + opt-in
  resolver round-trip.
- 163 total green across affected suites (T2 + existing empty-
  terminal mirror tests + finish_reason_tools_recovery +
  cascade_exhausted + internal_marker_stripping +
  tool_call_type_default + tool_use_enforcement + T3 + T1 persona
  + doctor persona probe).

## Composition

Per #143 sequencing:
- T3 / PR #145 (clarify default-out) — merged ✓
- T1 / PR #146 (persona default-flip) — merged ✓
- **T2 / this PR (empty-content recovery removal, env-gated)**
- Later: T4-T6 (tool list / iteration cap / summary fallback)

The legacy escape hatch design choice (vs. full removal) lets
operators roll back per-deployment if a previously-recovered edge
case surfaces in the field. Diagnostic from #140
(HERMES_DIAG_RAW_CAPTURE) remains the observability tool to spot
any uncovered case.