fix(agent): config-driven intent-ack continuation for all api_modes (#27881)

[teknium1]

Summary

The Discord/gateway agent could end a turn after only stating intent — "I will start by running a health check..." — without executing the announced tool call, forcing the user to re-prompt at every step (#27881). A continuation guard that catches this and nudges the model to proceed already existed, but it was hard-gated to the codex_responses api_mode, so Gemini / Claude / OpenRouter turns never benefited. This PR makes the guard config-driven and (when opted in) api_mode-agnostic, while keeping the default behavior byte-identical.

Root cause: agent/conversation_loop.py only ran _looks_like_codex_intermediate_ack when api_mode == "codex_responses". The detector logic itself was always provider-agnostic — only the wiring was Codex-only.

Changes

• hermes_cli/config.py: new agent.intent_ack_continuation setting. "auto" (default) = historical Codex-only behavior; true/list opts every api_mode in; false disables. Mirrors the four-mode shape of agent.tool_use_enforcement.
• agent/agent_init.py: read the flag onto agent._intent_ack_continuation.
• agent/agent_runtime_helpers.py: intent_ack_continuation_mode() (off / codex_only / all) centralizes resolution; looks_like_codex_intermediate_ack() gains require_workspace (default True). The opted-in path drops the codebase/filesystem requirement so general autonomous workflows (server ops, deploys, API calls) are caught, not just coding tasks.
• agent/conversation_loop.py: gate on the resolved mode; pass require_workspace=(mode == "codex_only"). The 2-nudge-per-turn cap and the future-ack / action-verb / short-content / no-prior-tool guards are unchanged.
• run_agent.py: forwarder signature updated.
• tests/agent/test_intent_ack_continuation.py: 13 invariant tests.

Validation

	auto (default)	true (opt-in)
Codex codebase ack	continues (unchanged)	continues
Gemini/Claude "I'll check the server" (#27881 repro)	turn ends (byte-stable)	continues
Real final answer / brainstorm / post-tool turn	no nudge	no nudge

• tests/agent/test_intent_ack_continuation.py: 13 passed.
• tests/run_agent/test_run_agent_codex_responses.py + tests/agent/test_system_prompt.py: 82 passed (auto path proven unchanged).

Infographic

[github-actions]

🔎 Lint report: hermes/hermes-a5ecb08c vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 11603 on HEAD, 11605 on base (✅ -2)

🆕 New issues (1):

Rule	Count
invalid-assignment	1

First entries

tests/run_agent/test_credits_notices_toggle.py:76: [invalid-assignment] invalid-assignment: Object of type `None` is not assignable to attribute `_credits_session_start_micros` of type `int`

✅ Fixed issues (2):

Rule	Count
unresolved-attribute	2

First entries

run_agent.py:3014: [unresolved-attribute] unresolved-attribute: Object of type `Self@get_credits_spent_micros` has no attribute `_credits_session_start_micros`
tests/run_agent/test_credits_notices_toggle.py:76: [unresolved-attribute] unresolved-attribute: Unresolved attribute `_credits_session_start_micros` on type `AIAgent`

Unchanged: 6088 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[tonydwb]

Code Review Summary

Verdict: Approved

Well-structured config-driven generalization of the intent-ack continuation feature. The new intent_ack_continuation config option (default "auto") extends the historical codex-only behavior to all api_modes when explicitly opted in. The three-mode shape (off/codex_only/all) mirrors tool_use_enforcement, keeping the config surface consistent.

Key observations:

• The require_workspace parameter is cleanly threaded from config through intent_ack_continuation_mode to the detector.
• Guardrails (no tool call history, short content, action verb) still apply regardless of mode, preventing false positives on conversational replies.
• 160 lines of new tests cover mode resolution, detector behavior, and edge cases comprehensively.
• The forwarder in run_agent.py correctly passes through the new parameter.

---

Reviewed by Hermes Agent