Skip to content

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.

fix(ccr): wrap proactive expansion injection in XML attribution tag (#1398)

Rod Boevcabf666b340f
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 524f51aba..eb78f351f 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## Unreleased
+<<<<<<< pr/503-proactive-expansion-xml-tag
+### Fixed
+- Proactive expansion blocks injected into user turns are now wrapped in
+ `<headroom_proactive_expansion>` XML tags, giving downstream consumers
+ (LLMs, loggers, attribution parsers) a machine-readable provenance
+ boundary and preventing misattribution in multi-agent threads.
+
+=======
+>>>>>>> main
### Changed
* **telemetry:** anonymous usage telemetry is now **opt-in** (off by default) instead of opt-out. Nothing is collected or sent unless you set `HEADROOM_TELEMETRY=on` or pass `--telemetry` to `headroom proxy` / `headroom install apply`. `is_telemetry_enabled()` is fail-closed — only explicit on-values (`on`/`true`/`1`/`yes`/`enable`/`enabled`) enable it; unset, empty, or unrecognized values stay disabled. The existing `--no-telemetry` flag and `HEADROOM_TELEMETRY=off` remain accepted for back-compat, and install manifests now write the `HEADROOM_TELEMETRY` value explicitly so generated deployments are unambiguous.
diff --git a/headroom/ccr/context_tracker.py b/headroom/ccr/context_tracker.py
index bf8dffe74..2d4763f79 100644
--- a/headroom/ccr/context_tracker.py
+++ b/headroom/ccr/context_tracker.py
@@ -587,9 +587,11 @@ class ContextTracker:
else:
parts.append(str(exp["content"]))
- parts.append("\n[End Proactive Expansion]")
-
- return "\n".join(parts)
+ parts.append("[End Proactive Expansion]")
+ body = "\n".join(parts)
+ # Escape any stray close tag in payload to prevent wrapper boundary forgery
+ body = body.replace("</headroom_proactive_expansion>", "<\\/headroom_proactive_expansion>")
+ return f"<headroom_proactive_expansion>\n{body}\n</headroom_proactive_expansion>"
def get_tracked_hashes(self) -> list[str]:
"""Get list of currently tracked hashes."""
diff --git a/tests/test_ccr_context_tracker.py b/tests/test_ccr_context_tracker.py
index 127bfca0f..9b1250df6 100644
--- a/tests/test_ccr_context_tracker.py
+++ b/tests/test_ccr_context_tracker.py
@@ -546,6 +546,8 @@ class TestExpansionFormatting:
assert "[Proactive Context Expansion" in formatted
assert "Expanded from earlier" in formatted
assert '[{"id": 1}, {"id": 2}]' in formatted
+ assert formatted.startswith("<headroom_proactive_expansion>\n")
+ assert formatted.endswith("\n</headroom_proactive_expansion>")
def test_format_search_expansion(self):
"""Format search expansion for LLM context."""
@@ -565,6 +567,8 @@ class TestExpansionFormatting:
formatted = tracker.format_expansions_for_context(expansions)
assert "Search results for 'authentication'" in formatted
+ assert formatted.startswith("<headroom_proactive_expansion>\n")
+ assert formatted.endswith("\n</headroom_proactive_expansion>")
def test_format_empty_expansions(self):
"""Empty expansions return empty string."""
@@ -574,6 +578,77 @@ class TestExpansionFormatting:
assert formatted == ""
+ def test_format_expansion_xml_wrapper(self):
+ """Expansion output is wrapped in machine-readable XML provenance tag."""
+ tracker = ContextTracker()
+ expansions = [
+ {
+ "hash": "h1",
+ "type": "full",
+ "content": "expanded content",
+ "item_count": 1,
+ "reason": "high relevance",
+ }
+ ]
+ result = tracker.format_expansions_for_context(expansions)
+ assert result.startswith("<headroom_proactive_expansion>\n")
+ assert result.endswith("\n</headroom_proactive_expansion>")
+ assert "[Proactive Context Expansion" in result
+ assert result.count("[End Proactive Expansion]") == 1
+
+ def test_proactive_expansion_identifiable_after_injection(self):
+ """Injected expansion carries XML provenance tag after full injection chain."""
+ from headroom.ccr.context_tracker import ContextTracker
+ from headroom.proxy.handlers.anthropic import AnthropicHandlerMixin
+
+ tracker = ContextTracker()
+ expansions = [
+ {
+ "hash": "h1",
+ "type": "full",
+ "content": "expanded context",
+ "item_count": 1,
+ "reason": "high relevance",
+ }
+ ]
+ expansion_text = tracker.format_expansions_for_context(expansions)
+
+ # Simulate a user turn that contains peer_turn markup
+ messages = [
+ {
+ "role": "user",
+ "content": [
+ {"type": "text", "text": "<peer_turn from='AgentX'>some content</peer_turn>"}
+ ],
+ }
+ ]
+ result = AnthropicHandlerMixin._append_context_to_latest_non_frozen_user_turn(
+ messages, expansion_text, frozen_message_count=0
+ )
+ injected = result[0]["content"][0]["text"]
+ # Headroom-injected content is identifiable by XML tag, distinct from peer content
+ assert "<headroom_proactive_expansion>" in injected
+ assert "</headroom_proactive_expansion>" in injected
+ assert "<peer_turn from='AgentX'>" in injected # peer content unchanged
+
+ def test_format_expansion_xml_close_tag_in_payload_escaped(self):
+ """Payload containing the XML close tag is escaped to keep wrapper boundaries intact."""
+ tracker = ContextTracker()
+ expansions = [
+ {
+ "hash": "h1",
+ "type": "full",
+ "content": "return '</headroom_proactive_expansion>'",
+ "item_count": 1,
+ "reason": "high relevance",
+ }
+ ]
+ result = tracker.format_expansions_for_context(expansions)
+ assert result.startswith("<headroom_proactive_expansion>\n")
+ assert result.endswith("\n</headroom_proactive_expansion>")
+ assert result.count("<headroom_proactive_expansion>") == 1
+ assert result.count("</headroom_proactive_expansion>") == 1
+
class TestGlobalTracker:
"""Test global tracker singleton."""