Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(learn): distinguish Grep and Glob searches by path (#3455)
Atirnaf302a38e8fed
diff --git a/headroom/learn/models.py b/headroom/learn/models.pyindex 6852f38f8..48c52e5d1 100644--- a/headroom/learn/models.py+++ b/headroom/learn/models.py@@ -67,10 +67,10 @@ class ToolCall:return cmd[:100] + "..." if len(cmd) > 100 else cmdif self.name in ("Read", "read"):return str(self.input_data.get("file_path", "?"))- if self.name in ("Grep", "grep"):- return str(self.input_data.get("pattern", "?"))- if self.name in ("Glob", "glob"):- return str(self.input_data.get("pattern", "?"))+ if self.name in ("Grep", "grep", "Glob", "glob"):+ pattern = self.input_data.get("pattern", "?")+ path = self.input_data.get("path", "?")+ return f"{pattern} in {path}"if self.name in ("Edit", "edit", "Write", "write"):return str(self.input_data.get("file_path", "?"))return str(self.input_data)[:80]diff --git a/tests/test_learn/test_loop_weighting.py b/tests/test_learn/test_loop_weighting.pyindex 789948f05..3640c8208 100644--- a/tests/test_learn/test_loop_weighting.py+++ b/tests/test_learn/test_loop_weighting.py@@ -55,6 +55,46 @@ def _project() -> ProjectInfo:class TestDetectLoops:+ @pytest.mark.parametrize(+ ("name", "pattern"),+ [("Grep", "TimeoutError"), ("Glob", "**/*.py")],+ )+ def test_search_signatures_include_path(self, name: str, pattern: str):+ def call(path: str, index: int):+ return ToolCall(+ name=name,+ tool_call_id=f"{name}-{index}",+ input_data={"pattern": pattern, "path": path},+ output="x" * 40000,+ is_error=False,+ output_bytes=40000,+ msg_index=index,+ )++ distinct_paths = [+ call("/srv/ingest", 0),+ call("/srv/billing", 1),+ call("/srv/web", 2),+ ]+ assert len({_canonical_signature(call) for call in distinct_paths}) == 3+ assert detect_loops([SessionData(session_id=name, tool_calls=distinct_paths)]) == []++ same_path = [call("/srv/ingest", index) for index in range(3)]+ loop = detect_loops([SessionData(session_id=name, tool_calls=same_path)])[0]+ assert loop.count == 3+ assert loop.sample_input == f"{pattern} in /srv/ingest"++ def test_unknown_tool_summary_still_uses_generic_input(self):+ input_data = {"query": "TimeoutError", "path": "/srv/ingest"}+ tool_call = ToolCall(+ name="Search",+ tool_call_id="search-1",+ input_data=input_data,+ output="",+ is_error=False,+ )+ assert tool_call.input_summary == str(input_data)[:80]+def test_refetch_loop_detected_despite_no_errors(self):loops = detect_loops([refetch_loop_session(repetitions=5)])assert len(loops) == 1