[FIX] Double open_kitchen Context Cost — Pre-Reveal / Deferred-Recall Architectural Immunity (#4092)

## Summary

The `open_kitchen` MCP tool lacks a unified gate-infrastructure /
content-delivery separation. When the Codex backend pre-reveals kitchen
tools at lifespan boot (`_pre_reveal_kitchen`), it opens the gate and
performs side effects (hook config, quota prime, registry) but does NOT
set `ctx.recipe_name`. When the agent subsequently calls
`open_kitchen(name="X")`, the `_is_deferred_recall` guard fails
(`recipe_name == ""` != `"X"`), causing the full
`_open_kitchen_handler()` to re-execute — redundantly re-opening the
gate, generating a new `kitchen_id`, re-writing hook config, re-priming
quota, and restarting the quota refresh loop. Then the recipe is loaded
and the full ~100KB payload is returned. The pre-reveal work is wasted
and the gate infrastructure is initialized twice.

Additionally, the `_is_deferred_recall` path (lines 595-683 of
`tools_kitchen.py`) returns the **full** `LoadRecipeResult` including
the `content` field (~72-102KB recipe YAML), with no `ingredients_only`
stripping applied. This means every deferred recall re-sends the entire
recipe.

The architectural weakness is that `open_kitchen` conflates three
concerns into a single code path: (1) gate infrastructure setup, (2)
recipe content loading, and (3) response payload assembly. The
`_is_deferred_recall` flag is a boolean that either skips or runs the
handler wholesale, with no intermediate state for "gate already open,
just load recipe."

Closes #4089

## Implementation Plan

Plan file:
`/home/talon/projects/autoskillit-runs/remediation-20260612-091855-490847/.autoskillit/temp/rectify/rectify_double_open_kitchen_context_cost_2026-06-12_091855.md`

🤖 Generated with [Claude Code](https://claude.com/claude-code) via
AutoSkillit
<!-- autoskillit:pipeline-signature
steps=prepare_pr,run_arch_lenses,compose_pr,annotate_pr_diff,review_pr
-->

## Token Usage Summary

| Step | Model | count | uncached | output | cache_read | peak_ctx |
turns | cache_write | time |

|------|-------|-------|----------|--------|------------|----------|-------|-------------|------|
| rectify* | opus[1m] | 1 | 7.3k | 20.6k | 2.7M | 120.6k | 72 | 99.9k |
25m 37s |
| review_approach* | opus[1m] | 1 | 24 | 6.4k | 222.3k | 52.6k | 12 |
35.0k | 4m 54s |
| dry_walkthrough* | opus | 1 | 40 | 11.9k | 890.7k | 95.9k | 36 | 96.1k
| 6m 50s |
| implement* | MiniMax-M3 | 1 | 5.4M | 14.7k | 0 | 0 | 115 | 0 | 13m 16s
|
| audit_impl* | opus[1m] | 1 | 34 | 14.4k | 280.3k | 63.5k | 20 | 48.7k
| 9m 2s |
| prepare_pr* | MiniMax-M3 | 1 | 292.3k | 3.3k | 0 | 0 | 21 | 0 | 1m 7s
|
| compose_pr* | MiniMax-M3 | 1 | 348.4k | 1.3k | 0 | 0 | 14 | 0 | 58s |
| **Total** | | | 6.0M | 72.6k | 4.1M | 120.6k | | 279.7k | 1h 1m |

\* *Step used a non-Anthropic provider; caching behavior may differ.*

## Token Efficiency

| Step | LoC Changed | cache_read/LoC | cache_write/LoC | output/LoC |
|------|-------------|----------------|-----------------|------------|
| rectify | 0 | — | — | — |
| review_approach | 0 | — | — | — |
| dry_walkthrough | 0 | — | — | — |
| implement | 280 | 0.0 | 0.0 | 52.5 |
| audit_impl | 0 | — | — | — |
| prepare_pr | 0 | — | — | — |
| compose_pr | 0 | — | — | — |
| **Total** | **280** | 14712.9 | 999.1 | 259.1 |

## Model Usage Breakdown

| Model | steps | uncached | output | cache_read | cache_write | time |
|-------|-------|----------|--------|------------|-------------|------|
| opus[1m] | 3 | 7.4k | 41.4k | 3.2M | 183.7k | 39m 33s |
| opus | 1 | 40 | 11.9k | 890.7k | 96.1k | 6m 50s |
| MiniMax-M3 | 3 | 6.0M | 19.3k | 0 | 0 | 15m 21s |

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Trecek

src/autoskillit/pipeline/context.py

@@ -168,6 +168,7 @@ class ToolContext:
     recipe_content_hash: str = field(default="")
     recipe_composite_hash: str = field(default="")
     recipe_version: str = field(default="")
+    gate_infrastructure_ready: bool = field(default=False)
     kitchen_id: str = field(default="")
     active_recipe_packs: frozenset[str] | None = field(default_factory=lambda: None)
     active_recipe_features: frozenset[str] | None = field(default_factory=lambda: None)

src/autoskillit/server/_lifespan.py

@@ -342,6 +342,7 @@ async def _pre_reveal_kitchen(ctx: Any) -> None:
     register_active_kitchen(ctx.kitchen_id, os.getpid(), str(ctx.project_dir))
     _write_hook_config()
     await _prime_quota_cache()
+    ctx.gate_infrastructure_ready = True
 
 
 async def _food_truck_auto_gate_boot(ctx: Any) -> None:

src/autoskillit/server/tools/tools_kitchen.py

@@ -276,6 +276,7 @@ async def _open_kitchen_handler() -> str | None:
     except Exception:
         logger.warning("open_kitchen_registry_failed", exc_info=True)
 
+    ctx.gate_infrastructure_ready = True
     return None
 
 
@@ -331,6 +332,7 @@ def _close_kitchen_handler() -> None:
     ctx.recipe_content_hash = ""
     ctx.recipe_composite_hash = ""
     ctx.recipe_version = ""
+    ctx.gate_infrastructure_ready = False
     logger.info("close_kitchen", gate_state="closed")
     if (log := ctx.github_api_log) is not None:
         orphan_usage = log.drain(ctx.kitchen_id)
@@ -515,15 +517,30 @@ async def open_kitchen(
         disabled_subsets = _get_ctx().config.subsets.disabled
 
         _ctx_pre = _get_ctx()
-        _is_deferred_recall = (
-            name is not None and _ctx_pre.gate.enabled and _ctx_pre.recipe_name == name
-        )
+        _skip_handler = _ctx_pre.gate_infrastructure_ready
+        tool_ctx = _get_ctx()
 
-        if not _is_deferred_recall:
+        if not _skip_handler:
             handler_err = await _open_kitchen_handler()
             if handler_err is not None:
                 return handler_err
+        else:
+            _ctx_post = _get_ctx()
+            if _ctx_post.quota_refresh_task is None:
+                try:
+                    _ctx_post.quota_refresh_task = create_background_task(
+                        _quota_refresh_loop(
+                            _ctx_post.config.quota_guard,
+                            provider=resolve_provider(_ctx_post.config.providers.default_provider),
+                        ),
+                        label="quota_refresh_loop",
+                    )
+                except Exception:
+                    logger.warning(
+                        "open_kitchen_quota_refresh_deferred_start_failed", exc_info=True
+                    )
 
+        if not _skip_handler:
             _kctx_pre = _get_ctx()
             _skip_notify = (
                 _kctx_pre.backend is not None
@@ -538,6 +555,7 @@ async def open_kitchen(
                     logger.warning(
                         "open_kitchen_failure", stage="enable_components", exc_info=True
                     )
+                    tool_ctx.gate_infrastructure_ready = False
                     return _kitchen_failure_envelope(exc, stage="enable_components")
 
             try:
@@ -550,8 +568,16 @@ async def open_kitchen(
                 )
             except Exception as exc:
                 logger.warning("open_kitchen_failure", stage="redisable_subsets", exc_info=True)
+                tool_ctx.gate_infrastructure_ready = False
                 return _kitchen_failure_envelope(exc, stage="redisable_subsets")
 
+        _is_deferred_recall = (
+            name is not None
+            and _ctx_pre.gate.enabled
+            and _ctx_pre.recipe_name == name
+            and _ctx_pre.recipe_name != ""
+        )
+
         _forbidden_list = ", ".join(PIPELINE_FORBIDDEN_TOOLS)
         _ctx = _get_ctx()
         _categories = _build_tool_category_listing(
@@ -645,6 +671,7 @@ async def open_kitchen(
                 # Default to False for missing 'valid' so a absent key is treated as invalid
                 if not result.get("valid", False) or not result.get("content", ""):
                     tool_ctx.gate.disable()
+                    tool_ctx.gate_infrastructure_ready = False
                     return _recipe_validation_error_response(name, result)
                 # Dispatch-feasibility preflight: verify the backend can enforce
                 # all fix-required hooks for the recipe's run_skill steps.
@@ -658,6 +685,7 @@ async def open_kitchen(
                     )
                     if _preflight_err is not None:
                         tool_ctx.gate.disable()
+                        tool_ctx.gate_infrastructure_ready = False
                         await ctx.disable_components(tags={"kitchen"})
                         return _preflight_err
                 result["success"] = True
@@ -680,6 +708,8 @@ async def open_kitchen(
                     )
                     if _override_warnings:
                         result["warnings"] = _override_warnings
+                if ingredients_only:
+                    result = strip_ingredients_only_keys(result)
                 return json.dumps(result)
             try:
                 result = tool_ctx.recipes.load_and_validate(
@@ -749,6 +779,7 @@ async def open_kitchen(
 
             if not result.get("valid", False) or not result.get("content", ""):
                 tool_ctx.gate.disable()
+                tool_ctx.gate_infrastructure_ready = False
                 return _recipe_validation_error_response(name, result)
 
             # Dispatch-feasibility preflight: verify the backend can enforce
@@ -763,6 +794,7 @@ async def open_kitchen(
                 )
                 if _preflight_err is not None:
                     tool_ctx.gate.disable()
+                    tool_ctx.gate_infrastructure_ready = False
                     await ctx.disable_components(tags={"kitchen"})
                     return _preflight_err
 

tests/arch/test_subpackage_isolation.py

@@ -962,14 +962,16 @@ def test_data_directories_are_not_python_packages() -> None:
         "co-located with the execution engine that calls them",
     ),
     "tools_kitchen.py": (
-        1200,
+        1250,
         "REQ-CNST-010-E7: kitchen tool handlers — open_kitchen and lock_ingredients require "
         "inline validation helpers (_check_override_keys, _build_ingredient_key_suggestions) "
         "for ingredient key validation; splitting would cross import-layer boundaries; "
         "backend capability promotion delegated to _promote_capability_keys in _auto_overrides; "
         "fail-closed validity gate on both deferred-recall and normal paths adds structural "
         "error propagation from LoadRecipeResult; get_recipe validity guard adds 9 lines; "
-        "dispatch-feasibility preflight wiring on both paths with gate-close on failure",
+        "dispatch-feasibility preflight wiring on both paths with gate-close on failure; "
+        "gate_infrastructure_ready guard restructuring adds 22 lines for handler-skip path, "
+        "quota_refresh_loop deferred start, and all four gate.disable() rollback resets",
     ),
     "tools_execution.py": (
         1130,

tests/infra/test_schema_version_convention.py

@@ -122,8 +122,8 @@ def _is_yaml_dump(node: ast.expr) -> bool:
     ("src/autoskillit/server/tools/tools_kitchen.py", 171),
     ("src/autoskillit/server/tools/tools_kitchen.py", 190),
     ("src/autoskillit/server/tools/tools_kitchen.py", 224),
-    ("src/autoskillit/server/tools/tools_kitchen.py", 918),
-    ("src/autoskillit/server/tools/tools_kitchen.py", 978),
+    ("src/autoskillit/server/tools/tools_kitchen.py", 950),
+    ("src/autoskillit/server/tools/tools_kitchen.py", 1010),
     # tools_pipeline_tracker.py — tracker_data dict
     ("src/autoskillit/server/tools/tools_pipeline_tracker.py", 166),
     # tools_status.py — mcp_data dict

tests/server/conftest.py

@@ -107,6 +107,7 @@ def _make_mock_ctx() -> MagicMock:
     ctx.project_dir = Path("/fake/project")
     ctx.config.subsets.disabled = []  # REQ-VIS-008: no subsets disabled by default
     ctx.active_recipe_ingredients = None
+    ctx.gate_infrastructure_ready = False
     return ctx
 
 

tests/server/test_kitchen_lifecycle.py

@@ -87,3 +87,42 @@ async def test_close_kitchen_removes_tracker_dir(monkeypatch, tmp_path):
         _close_kitchen_handler()
 
     assert not tracker_dir.exists()
+
+
+async def test_back_to_back_open_close_open_resets_infrastructure(monkeypatch, tmp_path):
+    """After close, gate_infrastructure_ready must be False so a re-open runs the handler."""
+    monkeypatch.chdir(tmp_path)
+
+    ctx = make_context(
+        AutomationConfig(),
+        runner=None,
+        plugin_source=DirectInstall(plugin_dir=tmp_path),
+        project_dir=tmp_path,
+    )
+    monkeypatch.setattr(_state, "_ctx", ctx)
+    monkeypatch.setattr(_state, "_startup_ready", None)
+
+    with (
+        patch("autoskillit.server.tools.tools_kitchen._prime_quota_cache", new_callable=AsyncMock),
+        patch("autoskillit.core.register_active_kitchen"),
+        patch("autoskillit.core.unregister_active_kitchen"),
+    ):
+        result1 = await _open_kitchen_handler()
+        assert result1 is None
+        assert ctx.gate_infrastructure_ready is True
+        first_task = ctx.quota_refresh_task
+
+        _close_kitchen_handler()
+        assert ctx.gate_infrastructure_ready is False
+        await asyncio.sleep(0)
+        assert first_task.cancelled() or first_task.done()
+
+        result2 = await _open_kitchen_handler()
+        assert result2 is None
+        assert ctx.gate_infrastructure_ready is True
+        second_task = ctx.quota_refresh_task
+        assert second_task is not first_task
+
+        _close_kitchen_handler()
+        await asyncio.sleep(0)
+        assert second_task.cancelled() or second_task.done()

tests/server/test_lifespan_skill_boot.py

@@ -353,6 +353,9 @@ async def test_codex_non_headless_pre_reveal_opens_gate(self, build_ctx, monkeyp
         assert ctx.kitchen_id is not None, "kitchen_id should be set by pre-reveal path"
         assert ctx.active_recipe_packs == frozenset()
         assert ctx.active_recipe_steps == {}
+        assert ctx.gate_infrastructure_ready is True, (
+            "gate_infrastructure_ready should be True after pre-reveal"
+        )
 
     @pytest.mark.anyio
     async def test_codex_non_headless_pre_reveal_suppresses_disabled_subset(