fix(OMN-15959): delete zombie handler_llm_cli_subprocess.py from node_llm_inference_effect

.github/workflows/hostile-reviewer.yml

@@ -5,9 +5,13 @@
 #
 # Adversarial multi-model code review gate on every pull request (OMN-8468).
 #
-# Invokes omniintelligence.review_pairing.cli_review (node-based path via
-# HandlerLlmCliSubprocess / adapter_ai_reviewer) with the two currently-live
-# local models (qwen3-review on the 5090, qwen3-review-b on the 4090 --
+# Invokes omniintelligence.review_pairing.cli_review directly (a standalone
+# module calling model endpoints over HTTP; it does not import or route
+# through any omnibase_infra handler -- OMN-15959 corrected this stale
+# doc-comment, which named a since-deleted HandlerLlmCliSubprocess class that
+# omniintelligence.review_pairing.cli_review never actually imported) with
+# the two currently-live local models (qwen3-review on the 5090,
+# qwen3-review-b on the 4090 --
 # OMN-14176; deepseek-r1 no longer serves on the local fleet). Codex is
 # excluded from the initial pilot -- add --model codex once non-interactive
 # OAuth auth is confirmed on the omnibase-ci runner (OMN-8524).
@@ -407,7 +411,7 @@ jobs:
               '| `blocked` | CRITICAL findings found | **Yes** |',
               '| `degraded` | All models unavailable (infra) | No (pilot) |',
               '',
-              '_Powered by [omniintelligence.review_pairing.cli_review](https://github.com/OmniNode-ai/omniintelligence) — node-based adversarial review via HandlerLlmCliSubprocess (OMN-8468/OMN-8524)_',
+              '_Powered by [omniintelligence.review_pairing.cli_review](https://github.com/OmniNode-ai/omniintelligence) — multi-model adversarial review (OMN-8468/OMN-8524)_',
             ].join('\n');
 
             const { data: comments } = await github.rest.issues.listComments({

docs/evidence/OMN-12548/determinism-audit.json

@@ -1,7 +1,7 @@
 {
   "epic": "OMN-12525",
   "gate_posture": "Report-only at S0 (design D3). Each entry is a real multi-match tie resolved by registration/insertion order \u2014 a latent production hazard ticketed under OMN-12525, NOT gate-blocking in this parity PR.",
-  "generated_at_utc": "2026-08-20T13:40:04.122324+00:00",
+  "generated_at_utc": "2026-08-22T21:25:19.434279+00:00",
   "order_sensitive_count": 27,
   "order_sensitive_probes": [
     {

scripts/ci/infra-node-allowlist.txt

@@ -127,7 +127,6 @@ src/omnibase_infra/nodes/node_llm_inference_effect/handlers/bifrost/model_bifros
 src/omnibase_infra/nodes/node_llm_inference_effect/handlers/bifrost/model_bifrost_request.py  # approved existing infra handler path; new business-domain handlers belong in omnimarket.
 src/omnibase_infra/nodes/node_llm_inference_effect/handlers/bifrost/model_bifrost_response.py  # approved existing infra handler path; new business-domain handlers belong in omnimarket.
 src/omnibase_infra/nodes/node_llm_inference_effect/handlers/bifrost/model_bifrost_routing_rule.py  # approved existing infra handler path; new business-domain handlers belong in omnimarket.
-src/omnibase_infra/nodes/node_llm_inference_effect/handlers/handler_llm_cli_subprocess.py  # approved existing infra handler path; new business-domain handlers belong in omnimarket.
 src/omnibase_infra/nodes/node_llm_inference_effect/handlers/handler_llm_inference_command.py  # infra-owned runtime command adapter for the LLM inference effect; business routing remains outside this handler.
 src/omnibase_infra/nodes/node_llm_inference_effect/handlers/handler_llm_openai_compatible.py  # approved existing infra handler path; new business-domain handlers belong in omnimarket.
 src/omnibase_infra/nodes/node_merge_gate_effect/handlers/__init__.py  # approved existing infra handler path; new business-domain handlers belong in omnimarket.

src/omnibase_infra/models/coding_agent/enum_cli_backend_status.py

@@ -2,11 +2,12 @@
 # SPDX-License-Identifier: MIT
 """Structured CLI subprocess failure classes for coding-agent invocations.
 
-Mirrors the failure vocabulary in
-``node_llm_inference_effect.handlers.handler_llm_cli_subprocess.EnumCliBackendStatus``.
-Phase D (OMN-13250) collapses the duplicate by relocating the inference handler;
-until then the coding-agent surface owns its own copy so the invoke effect does
-not import a sibling node's private handler module.
+OMN-15959: the sibling copy of this vocabulary in
+``node_llm_inference_effect.handlers.handler_llm_cli_subprocess`` was deleted
+(OMN-13250's Phase D DoD claimed that collapse had already happened; it had
+not — the handler was still live with 40 tests). This module is now the sole
+owner of the CLI-backend failure vocabulary; the coding-agent surface does not
+import a sibling node's private handler module.
 """
 
 from __future__ import annotations

src/omnibase_infra/nodes/node_coding_agent_invoke_effect/contract.yaml

@@ -37,9 +37,9 @@
 # transport-envelope-shaped output events before publish, preserving the
 # contract-declared terminal event semantics without node-local envelope code.
 name: node_coding_agent_invoke_effect
-contract_version: {major: 1, minor: 0, patch: 1}
+contract_version: {major: 1, minor: 0, patch: 2}
 node_type: EFFECT_GENERIC
-node_version: {major: 1, minor: 0, patch: 1}
+node_version: {major: 1, minor: 0, patch: 2}
 description: >
   Coding-agent invoke effect. Runs the claude/codex CLI subprocess in the target workspace and returns a system-derived ModelCodingAgentResult: status, exit code, git-derived files_changed + diff (captured AFTER the subprocess exits, never trusted from agent stdout), diff_hash, starting_head_sha, structured error_class (UNAVAILABLE/TIMEOUT/SUBPROCESS_ERROR/EMPTY_RESPONSE), timed_out, and duration. The handler uses the canonical def-B runtime shape: the contract event_model payload is validated by the shared runtime adapter before handle(command), and the bare ModelCodingAgentResult return is wrapped for EFFECT publication by the runtime. shutil.which gates availability; the subprocess spawns in its own process group and a timeout kills the whole group. The subprocess + git I/O are injected seams so this node is the single sanctioned model-CLI shell-out surface. agent-reported output/usage are advisory only.
 

src/omnibase_infra/nodes/node_coding_agent_invoke_effect/handlers/handler_coding_agent_invoke.py

@@ -2,10 +2,11 @@
 # SPDX-License-Identifier: MIT
 """Coding-agent invoke EFFECT handler (OMN-13247, plan §5.1 / §5.7).
 
-The ONLY I/O node in the workflow. Ported from
-``node_llm_inference_effect.handlers.handler_llm_cli_subprocess`` and upgraded
-from read-only single-shot inference to a workspace-write agentic invocation with
-git-derived diff capture:
+The ONLY I/O node in the workflow. Originally ported from
+``node_llm_inference_effect.handlers.handler_llm_cli_subprocess`` (deleted
+under OMN-15959 — that handler's shell-out architecture was superseded by this
+node, not merely duplicated) and upgraded from read-only single-shot inference
+to a workspace-write agentic invocation with git-derived diff capture:
 
   - ``shutil.which`` availability gate (UNAVAILABLE if the CLI is missing).
   - structured failure classes (UNAVAILABLE / TIMEOUT / SUBPROCESS_ERROR /

src/omnibase_infra/nodes/node_llm_inference_effect/contract.yaml

@@ -1,27 +1,24 @@
 # SPDX-FileCopyrightText: 2025 OmniNode.ai Inc.
 # SPDX-License-Identifier: MIT
 # OMN-8735: handler constructor compliance refactor (2026-04-15)
-# OMN-8735 follow-up: HandlerLlmCliSubprocess null-guard compliance (2026-04-16)
 # OMN-12816: additive extra_body passthrough — provider-specific request-body
 #   fields (e.g. chat_template_kwargs.enable_thinking) merged in _build_payload
 #   without overriding declared fields (2026-06-07)
-# OMN-13158: HandlerLlmCliSubprocess resolves OAuth/subscription CLI backends
-#   from request.model (claude-cli, gemini-cli, opencode-cli).
-# OMN-13215: the codex-cli shell-execution path was REMOVED. The delegation
-#   ceiling executes over the canonical HTTP inference path; no codex subprocess
-#   operation or backend remains.
 # OMN-13204: HandlerLlmOpenaiCompatible Protocol Conformance Note corrected to
 #   the real protocol I/O types — ProtocolHandler.execute(ModelProtocolRequest)
 #   -> ModelProtocolResponse and ProtocolMessageHandler.handle(ModelEventEnvelope)
 #   -> ModelHandlerOutput — replacing the deleted ModelOnexEnvelope reference
 #   (B5 residual of OMN-13196; deploy-gate runtime-path docstring edit, behavior
 #   unchanged). (2026-06-17)
-# OMN-14804: HandlerLlmCliSubprocess now exposes the canonical async handle()
-#   dispatch entrypoint for gemini/claude/opencode CLI operations; the retained
-#   execute_cli_inference accessor delegates to the same owned core. (2026-07-19)
 # OMN-15115: HandlerLlmOpenaiCompatible accepts per-request max_retries and the
 #   OpenAI-compatible request model permits timeout_seconds up to 1800s for slow
 #   local endpoints measured below the old 600s ceiling. (2026-07-25)
+# OMN-15959: HandlerLlmCliSubprocess and its inference.{gemini,claude,opencode}_cli
+#   shell-out operations were DELETED. OMN-13250 (Phase D) claimed this handler
+#   was already stripped to HTTP-only; it was not — the file, its 3 operations,
+#   and 40 tests were still live. The sanctioned CLI shell-out surface is
+#   node_coding_agent_invoke_effect (descriptor.agent_invocation: true) only.
+#   (2026-08-12)
 # Copyright (c) 2026 OmniNode Team
 #
 # ONEX Node Contract
@@ -36,8 +33,8 @@ contract_name: "node_llm_inference_effect"
 node_name: "node_llm_inference_effect"
 contract_version:
   major: 1
-  minor: 4
-  patch: 4
+  minor: 5
+  patch: 0
 node_version:
   major: 0
   minor: 1
@@ -87,27 +84,6 @@ handler_routing:
         module: "omnibase_infra.nodes.node_llm_inference_effect.handlers.handler_llm_openai_compatible"
         handler_type: "HTTP"
       description: "Execute LLM inference via OpenAI-compatible API (vLLM, OpenAI, GLM/Z.AI, etc.) with per-request timeout_seconds up to 1800s and max_retries passthrough."
-    # Gemini CLI Subprocess Handler
-    - operation: "inference.gemini_cli"
-      handler:
-        name: "HandlerLlmCliSubprocess"
-        module: "omnibase_infra.nodes.node_llm_inference_effect.handlers.handler_llm_cli_subprocess"
-      description: "Execute LLM inference via Gemini CLI subprocess (gemini -p)"
-    # OMN-13215: the Codex CLI subprocess operation (inference.codex_cli) was
-    # REMOVED. The delegation ceiling executes over the canonical HTTP inference
-    # path (HandlerLlmOpenaiCompatible) — no codex shell-out remains.
-    # Claude CLI Subprocess Handler (OMN-10137)
-    - operation: "inference.claude_cli"
-      handler:
-        name: "HandlerLlmCliSubprocess"
-        module: "omnibase_infra.nodes.node_llm_inference_effect.handlers.handler_llm_cli_subprocess"
-      description: "Execute LLM inference via Claude CLI subprocess (claude -p --output-format json)"
-    # opencode CLI Subprocess Handler (OMN-10137)
-    - operation: "inference.opencode_cli"
-      handler:
-        name: "HandlerLlmCliSubprocess"
-        module: "omnibase_infra.nodes.node_llm_inference_effect.handlers.handler_llm_cli_subprocess"
-      description: "Execute LLM inference via opencode CLI subprocess (opencode run --format json --pure)"
   execution_mode: "single"
   max_timeout_seconds: 1800
 # =============================================================================
@@ -224,7 +200,7 @@ metadata:
   author: "OmniNode Team"
   license: "MIT"
   created: "2026-02-13"
-  updated: "2026-07-19"
+  updated: "2026-08-12"
   tags:
     - effect
     - llm