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
Extract diagnostic decode policy (#1981)
JD Davis7fb920908922
diff --git a/headroom/proxy/diagnostic_decode_policy.py b/headroom/proxy/diagnostic_decode_policy.pynew file mode 100644index 000000000..9dc595601--- /dev/null+++ b/headroom/proxy/diagnostic_decode_policy.py@@ -0,0 +1,17 @@+"""Lossy byte decoding policy for diagnostics and logs."""++from __future__ import annotations++import codecs+++def safe_decode_for_logging(raw: bytes, *, max_bytes: int | None = None) -> str:+ """Decode bytes to a string for log/diagnostic display only.++ Wire/protocol parsers should decode complete protocol frames strictly. This+ policy is for already-discarded diagnostics where replacement characters are+ preferable to failing the error-reporting path.+ """+ blob = raw[:max_bytes] if max_bytes is not None else raw+ decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")+ return decoder.decode(bytes(blob), final=True)diff --git a/headroom/proxy/helpers.py b/headroom/proxy/helpers.pyindex 7a415e882..d050dacca 100644--- a/headroom/proxy/helpers.py+++ b/headroom/proxy/helpers.py@@ -25,6 +25,7 @@ from typing import TYPE_CHECKING, Any, Literal, castfrom headroom import paths as _pathsfrom headroom._subprocess import runfrom headroom.proxy import (+ diagnostic_decode_policy,request_limit_policy,sse_byte_buffer_policy,wire_debug_format_policy,@@ -565,16 +566,7 @@ def safe_decode_for_logging(raw: bytes, *, max_bytes: int | None = None) -> str:Use ``parse_sse_events_from_byte_buffer`` for SSE parsing instead."""- blob = raw[:max_bytes] if max_bytes is not None else raw- # Decode incrementally and represent any invalid bytes as the- # Unicode replacement character (�). Implemented via the- # `codecs` incremental decoder so we never reach for the- # forbidden `errors="ignore"`/`errors="replace"` keyword in the- # SSE-bearing modules.- import codecs as _codecs-- decoder = _codecs.getincrementaldecoder("utf-8")(errors="replace")- return decoder.decode(bytes(blob), final=True)+ return diagnostic_decode_policy.safe_decode_for_logging(raw, max_bytes=max_bytes)def parse_sse_events_from_byte_buffer(diff --git a/tests/test_diagnostic_decode_policy.py b/tests/test_diagnostic_decode_policy.pynew file mode 100644index 000000000..1cded9a44--- /dev/null+++ b/tests/test_diagnostic_decode_policy.py@@ -0,0 +1,20 @@+from __future__ import annotations++from headroom.proxy.diagnostic_decode_policy import safe_decode_for_logging+from headroom.proxy.helpers import safe_decode_for_logging as helper_safe_decode_for_logging+++def test_safe_decode_for_logging_decodes_utf8() -> None:+ assert safe_decode_for_logging("hello \u2603".encode()) == "hello \u2603"+++def test_safe_decode_for_logging_replaces_invalid_bytes() -> None:+ assert safe_decode_for_logging(b"ok\xffdone") == "ok\ufffddone"+++def test_safe_decode_for_logging_honors_max_bytes_before_decoding() -> None:+ assert safe_decode_for_logging(b"abcdef", max_bytes=3) == "abc"+++def test_helpers_safe_decode_delegates_to_policy() -> None:+ assert helper_safe_decode_for_logging(b"ok\xffdone") == safe_decode_for_logging(b"ok\xffdone")