Merge remote-tracking branch 'origin/main' into ironrod-local-patches

Author: IronRod Ops

agent/chat_completion_helpers.py

@@ -3270,6 +3270,10 @@ def _managed_summary_call(request, callback, *, retry_count: int):
         # tool_call was summarized away; Responses API rejects that as
         # "No tool call found for function call output".
         api_messages = agent._sanitize_api_messages(api_messages)
+        # Same send-path vision eviction as the main loop (#89296).
+        from agent.context_compressor import evict_stale_outbound_tool_images
+
+        evict_stale_outbound_tool_images(api_messages)
 
         # Same safety net as the main loop: drop thinking-only assistant
         # turns so Anthropic-family providers don't 400 the summary call.

agent/context_compressor.py

@@ -1772,6 +1772,23 @@ def _retire_stale_tool_result_images(
     return pruned
 
 
+def evict_stale_outbound_tool_images(
+    api_messages: List[Dict[str, Any]],
+    keep_newest: int = _MAX_KEEP_TOOL_IMAGES,
+) -> int:
+    """Drop stale screenshot/vision payloads from the per-call API copy.
+
+    Compression's keep-newest pass only runs when prune/compress fires, and
+    the Anthropic adapter's screenshot eviction only sees nested
+    ``tool_result`` blocks. OpenAI-style ``image_url`` tool results
+    otherwise ride every subsequent request until a 413 forces the reactive
+    strip (#89286). Call this on the cloned ``api_messages`` list after
+    sanitization so older frames never leave the box (#89296). Do not pass
+    persisted history — the rewrite is send-path only.
+    """
+    return _retire_stale_tool_result_images(api_messages, keep_newest=keep_newest)
+
+
 def _truncate_tool_call_args_json(args: str, head_chars: int = 200) -> str:
     """Shrink long string values inside a tool-call arguments JSON blob while
     preserving JSON validity.

agent/conversation_loop.py

@@ -2743,6 +2743,13 @@ def run_conversation(
         # gated on context_compressor — so orphans from session loading or
         # manual message manipulation are always caught.
         api_messages = agent._sanitize_api_messages(api_messages)
+        # Send-path vision eviction (#89296): compression only strips stale
+        # screenshots when prune fires, and the Anthropic adapter's keep-window
+        # never sees OpenAI-style tool-result image_url parts. The per-call
+        # clone is rewritten in place; persisted history is untouched.
+        from agent.context_compressor import evict_stale_outbound_tool_images
+
+        evict_stale_outbound_tool_images(api_messages)
 
         # One-time repeated-heal escalation notice (#96870): if the sanitizer
         # above just crossed the per-session heal threshold, deliver the

agent/relay_llm.py

@@ -1119,6 +1119,19 @@ def _provider_request(
             for key, value in headers.items()
             if str(key).lower() not in _RELAY_INTERNAL_PROVIDER_HEADERS
         }
+    # Relay's managed-call trace header maps to ``extra_headers`` for known SDK
+    # adapters and custom requests that already use that container. Other
+    # native transports receive protocol kwargs directly and may reject a new
+    # SDK-only argument. Preserve non-trace middleware headers as before.
+    supports_extra_headers = (
+        _relay_protocol(metadata) is not None or "extra_headers" in original
+    )
+    if headers and not supports_extra_headers:
+        headers = {
+            key: value
+            for key, value in headers.items()
+            if str(key).lower() != "traceparent"
+        }
     if headers:
         final["extra_headers"] = {
             **dict(final.get("extra_headers") or {}),

agent/relay_runtime.py

@@ -343,8 +343,8 @@ def acquire(
 
                 if self._activation is None:
                     # Hermes only enters Relay's initialization path after an
-                    # explicit opt-in. Relay currently owns any subsequent ambient
-                    # layering; a future discovery=False API can make this exact.
+                    # explicit opt-in. Relay 0.8 no longer layers repository-local
+                    # configuration onto this explicitly selected payload.
                     _resolve_plugin_awaitable(relay.plugin.initialize(plugin_config))
             except Exception as exc:
                 self._activation = None

agent/relay_tools.py

@@ -21,6 +21,7 @@ def execute(
     callback: Callable[[dict[str, Any]], Any],
     *,
     session_id: str,
+    tool_call_id: str | None = None,
     metadata: dict[str, Any] | None = None,
 ) -> tuple[Any, dict[str, Any]]:
     """Run one tool call through Relay and return its final arguments."""
@@ -52,7 +53,7 @@ def guarded(final_args: dict[str, Any]) -> Any:
             raise
         raw_result["value"] = result
         raw_result["json"] = _jsonable(result)
-        return raw_result["json"]
+        return runtime.relay.ToolExecutionResult(raw_result["json"])
 
     try:
         managed = _run_awaitable(
@@ -64,6 +65,7 @@ def guarded(final_args: dict[str, Any]) -> Any:
                 invoke,
                 handle=parent,
                 metadata=_jsonable(metadata or {}),
+                tool_call_id=tool_call_id or None,
             )
         )
     except BaseException as exc:
@@ -85,11 +87,12 @@ def guarded(final_args: dict[str, Any]) -> Any:
             return raw_result["value"], observed_args
         raise
 
-    if "value" in raw_result and _json_equal(managed, raw_result["json"]):
+    managed_result = managed.result
+    if "value" in raw_result and _json_equal(managed_result, raw_result["json"]):
         return raw_result["value"], observed_args
-    if isinstance(managed, str):
-        return managed, observed_args
-    return json.dumps(_jsonable(managed), ensure_ascii=False), observed_args
+    if isinstance(managed_result, str):
+        return managed_result, observed_args
+    return json.dumps(_jsonable(managed_result), ensure_ascii=False), observed_args
 
 
 def _jsonable(value: Any) -> Any:

agent/search_policy.py

@@ -0,0 +1,30 @@
+"""Shared directory pruning policy for broad recursive scans.
+
+These names identify version-control internals, dependency trees, generated
+artifacts, caches, and backup copies that are not useful results for broad
+agent-facing discovery. Ordinary search callers may still target an explicit
+path; broad diagnostic probes should apply this policy to recursive walks.
+"""
+
+from __future__ import annotations
+
+
+# Keep this policy conservative and name-based so it works for local and remote
+# shell backends alike. The same set is used by context discovery and search
+# probes; adding a directory here protects every broad recursive consumer.
+SEARCH_PRUNE_DIR_NAMES = frozenset({
+    # Version-control internals.
+    ".git", ".hg", ".svn",
+    # Dependency and vendored trees.
+    "node_modules", "venv", ".venv", "site-packages", "dist-packages",
+    "vendor", "third_party",
+    # Generated/build output.
+    "build", "dist", "target", "out", "coverage",
+    ".next", ".turbo", ".parcel-cache", ".nuxt", ".svelte-kit",
+    # Python and package-manager caches.
+    "__pycache__", ".cache", ".Trash", ".tox", ".nox", ".mypy_cache",
+    ".pytest_cache", ".ruff_cache", ".npm", ".yarn", ".pnpm-store",
+    ".gradle", ".m2", ".nuget",
+    # Backup copies.
+    "backups", "backup", ".backups",
+})

agent/subdirectory_hints.py

@@ -21,6 +21,7 @@
 from typing import Dict, Any, Optional, Set
 
 from agent.prompt_builder import _read_text_with_timeout, _scan_context_content
+from agent.search_policy import SEARCH_PRUNE_DIR_NAMES
 
 logger = logging.getLogger(__name__)
 
@@ -47,17 +48,9 @@
 # Prevents scanning all the way to / for deeply nested paths.
 _MAX_ANCESTOR_WALK = 5
 
-# Directory names that never contain authoritative project context.
-# Backups, vendored deps, VCS internals, and caches routinely hold *copies* of
-# AGENTS.md; loading those duplicates real context and inflates the prompt.
-_EXCLUDED_DIR_NAMES = frozenset({
-    "node_modules", "venv", ".venv", "__pycache__",
-    ".git", ".hg", ".svn",
-    ".Trash", ".cache", ".tox", ".mypy_cache", ".pytest_cache",
-    "site-packages", "dist-packages",
-    "backups", "backup", ".backups",
-    "vendor", "third_party",
-})
+# Shared with broad recursive search probes so context discovery and search do
+# not drift into different dependency/cache/build trees.
+_EXCLUDED_DIR_NAMES = SEARCH_PRUNE_DIR_NAMES
 
 
 def _is_ancestor_or_same(a: Path, b: Path) -> bool:

agent/tool_executor.py

@@ -771,6 +771,7 @@ def _hermes_pipeline(relay_args: dict[str, Any]) -> Any:
         function_args,
         _hermes_pipeline,
         session_id=str(getattr(agent, "session_id", "") or ""),
+        tool_call_id=tool_call_id or None,
         metadata={
             "task_id": effective_task_id or "",
             "turn_id": getattr(agent, "_current_turn_id", "") or "",

agent/usage_pricing.py

@@ -1278,6 +1278,10 @@ def get_pricing_entry(
         )
     if route.provider == "openrouter":
         return _openrouter_pricing_entry(route)
+
+    bundled_entry = _lookup_official_docs_pricing(route)
+    if bundled_entry:
+        return bundled_entry
     if route.base_url:
         entry = _pricing_entry_from_metadata(
             fetch_endpoint_model_metadata(route.base_url, api_key=api_key or ""),
@@ -1287,7 +1291,7 @@ def get_pricing_entry(
         )
         if entry:
             return entry
-    return _lookup_official_docs_pricing(route)
+    return None
 
 
 def normalize_usage(