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(thinking): leave a thinking block with a null or non-string value unchanged (#4562)
HetCreepcc1eb5402aad
diff --git a/headroom/transforms/thinking_compactor.py b/headroom/transforms/thinking_compactor.pyindex 46fda82d9..c03afd2e7 100644--- a/headroom/transforms/thinking_compactor.py+++ b/headroom/transforms/thinking_compactor.py@@ -162,7 +162,10 @@ def compact_thinking_to_text(if not (isinstance(block, dict) and block.get("type") == "thinking"):new_content.append(block)continue- text = block.get("thinking", "")+ text = block.get("thinking")+ if not isinstance(text, str): # null/non-str: leave the block as it is+ new_content.append(block)+ continuewords = len(text.split())if words < min_words:new_content.append(block)diff --git a/tests/test_thinking_compactor_version_gate.py b/tests/test_thinking_compactor_version_gate.pyindex 07238dcfb..8df296b6d 100644--- a/tests/test_thinking_compactor_version_gate.py+++ b/tests/test_thinking_compactor_version_gate.py@@ -51,3 +51,23 @@ def test_date_suffix_is_not_read_as_minor_version() -> None:assert bills_prior_thinking("claude-sonnet-4-20250514") is False# Sanity: the real 4.6 id still bills.assert bills_prior_thinking("claude-sonnet-4-6-20260101") is True+++def test_compact_thinking_to_text_leaves_null_thinking_block_alone() -> None:+ """``"thinking": null`` must not raise (docstring: "Never raises")."""+ from headroom.transforms.thinking_compactor import compact_thinking_to_text++ class _K:+ def compress(self, text, **kwargs):+ return type("R", (), {"compressed": text[:10]})()++ null_block = {"type": "thinking", "thinking": None, "signature": "s"}+ msgs = [+ {"role": "user", "content": "q"},+ {"role": "assistant", "content": [null_block, {"type": "text", "text": "a"}]},+ {"role": "user", "content": "q2"},+ {"role": "assistant", "content": [{"type": "text", "text": "b"}]},+ ]+ out, stats = compact_thinking_to_text(msgs, kompress=_K())+ assert out[1]["content"][0] == null_block+ assert stats["blocks"] == 0