fix(integrations): filter CCR-dropped sentinel in test iteration

The PR8 marker injection appends a sentinel object
{"_ccr_dropped": "<<ccr:HASH N_rows_offloaded>>"} to the kept-items
array on the lossy path so the LLM sees the retrieval pointer in the
prompt. Tests that iterate compressed arrays via subscript access
(e["level"], r["status"], i["labels"], m["text"]) hit KeyError on
the sentinel because it doesn't share the record schema.

Same root cause as the test_quality_retention fixes in PR8 -- these
integration tests were left out of that pass.

Ship a public helper headroom.transforms.smart_crusher.strip_ccr_sentinels
so tests can use it cleanly: `for e in strip_ccr_sentinels(entries):`
and production callers iterating compressed output get a single
canonical filter instead of inlining the _ccr_dropped check.

The 7 previously-failing tests in PR #292 CI now pass:
  - langchain test_100_percent_errors_preserved_logs
  - langchain test_errors_preserved_with_many_errors
  - langchain test_search_results_with_query_term
  - mcp test_all_log_errors_preserved
  - mcp test_slack_significant_compression_with_content
  - mcp test_database_error_status_preserved
  - mcp test_github_bugs_partial_preservation

753 tests across the integration + transforms + retention suites pass
locally. Plugin manifests auto-bumped 0.13.3 -> 0.13.4 by the
sync-plugin-versions hook (unrelated to this fix).

.claude-plugin/marketplace.json

@@ -5,14 +5,14 @@
   },
   "metadata": {
     "description": "Headroom marketplace for Claude Code and GitHub Copilot CLI plugins.",
-    "version": "0.13.2"
+    "version": "0.13.4"
   },
   "plugins": [
     {
       "name": "headroom",
       "source": "./plugins/headroom-agent-hooks",
       "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
-      "version": "0.13.2",
+      "version": "0.13.4",
       "author": {
         "name": "Headroom Contributors",
         "url": "https://github.com/chopratejas/headroom"

.github/plugin/marketplace.json

@@ -5,14 +5,14 @@
   },
   "metadata": {
     "description": "Headroom marketplace for Claude Code and GitHub Copilot CLI plugins.",
-    "version": "0.13.2"
+    "version": "0.13.4"
   },
   "plugins": [
     {
       "name": "headroom",
       "source": "./plugins/headroom-agent-hooks",
       "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
-      "version": "0.13.2",
+      "version": "0.13.4",
       "author": {
         "name": "Headroom Contributors",
         "url": "https://github.com/chopratejas/headroom"

headroom/transforms/smart_crusher.py

@@ -42,6 +42,39 @@ from .base import Transform
 logger = logging.getLogger(__name__)
 
 
+# ─── CCR sentinel ─────────────────────────────────────────────────────────
+#
+# When SmartCrusher's lossy path drops rows, it appends a sentinel object
+# `{"_ccr_dropped": "<<ccr:HASH N_rows_offloaded>>"}` to the kept-items
+# array. The LLM sees this in the prompt and can ask for the original via
+# the CCR retrieval tool. Downstream consumers that iterate the array
+# expecting a uniform schema (e.g. `for e in entries: e["level"]`) need
+# to skip the sentinel — that's what `strip_ccr_sentinels` is for.
+
+CCR_SENTINEL_KEY = "_ccr_dropped"
+
+
+def is_ccr_sentinel(item: Any) -> bool:
+    """True if `item` is a CCR-dropped sentinel object."""
+    return isinstance(item, dict) and CCR_SENTINEL_KEY in item
+
+
+def strip_ccr_sentinels(items: Any) -> Any:
+    """Return `items` with any CCR-dropped sentinel objects filtered out.
+
+    Pass this through any iteration over a compressed array's contents
+    when your code expects a uniform-schema list of records. The sentinel
+    carries a `<<ccr:HASH ...>>` marker for the LLM and shouldn't be
+    confused for a record — it has only the `_ccr_dropped` key.
+
+    Non-list inputs pass through unchanged so callers can wrap whatever
+    `json.loads` returned without first checking the shape.
+    """
+    if not isinstance(items, list):
+        return items
+    return [x for x in items if not is_ccr_sentinel(x)]
+
+
 # ─── Public dataclasses ───────────────────────────────────────────────────
 
 

plugins/headroom-agent-hooks/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "headroom",
-  "version": "0.13.2",
+  "version": "0.13.4",
   "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
   "author": {
     "name": "Headroom Contributors",

plugins/headroom-agent-hooks/.github/plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "headroom",
-  "version": "0.13.2",
+  "version": "0.13.4",
   "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
   "author": {
     "name": "Headroom Contributors",

tests/test_integrations/langchain/test_evals.py

@@ -18,6 +18,7 @@ import pytest
 from headroom.config import SmartCrusherConfig
 from headroom.providers import OpenAIProvider
 from headroom.transforms import SmartCrusher
+from headroom.transforms.smart_crusher import strip_ccr_sentinels
 
 
 # Test fixtures for realistic data
@@ -190,8 +191,12 @@ class TestErrorPreservation:
         json_match = re.search(r"(\{.*\})", compressed_output, re.DOTALL)
         compressed_data = json.loads(json_match.group(1) if json_match else compressed_output)
 
-        # Count preserved errors
-        compressed_errors = [e for e in compressed_data["entries"] if e["level"] == "ERROR"]
+        # Count preserved errors. Strip CCR-dropped sentinel objects
+        # before iterating — they carry the retrieval marker for the LLM
+        # but don't share the entry schema.
+        compressed_errors = [
+            e for e in strip_ccr_sentinels(compressed_data["entries"]) if e["level"] == "ERROR"
+        ]
 
         # CRITICAL: 100% of errors must be preserved
         assert len(compressed_errors) == len(original_errors), (
@@ -226,7 +231,9 @@ class TestErrorPreservation:
         json_match = re.search(r"(\{.*\})", compressed_output, re.DOTALL)
         compressed_data = json.loads(json_match.group(1) if json_match else compressed_output)
 
-        compressed_errors = [e for e in compressed_data["entries"] if e["level"] == "ERROR"]
+        compressed_errors = [
+            e for e in strip_ccr_sentinels(compressed_data["entries"]) if e["level"] == "ERROR"
+        ]
 
         # Even with many errors, ALL must be preserved
         assert len(compressed_errors) == len(original_errors), (
@@ -314,7 +321,7 @@ class TestRelevancePreservation:
         # At least some high-relevance results should be preserved
         # (BM25 may not catch all without exact keyword matches)
         compressed_high_relevance = [
-            r for r in compressed_data["results"] if r["relevance_score"] > 0.8
+            r for r in strip_ccr_sentinels(compressed_data["results"]) if r["relevance_score"] > 0.8
         ]
 
         # With BM25, we should preserve at least 1 high-relevance result

tests/test_integrations/mcp/test_server.py

@@ -19,6 +19,7 @@ from headroom.integrations.mcp import (
     compress_tool_result_with_metrics,
 )
 from headroom.providers import OpenAIProvider
+from headroom.transforms.smart_crusher import strip_ccr_sentinels
 
 # ============================================================================
 # Test Fixtures
@@ -280,7 +281,9 @@ class TestMCPErrorPreservation:
 
         compressed_data = json.loads(result.compressed_content)
         compressed_errors = [
-            e for e in compressed_data["entries"] if e["level"] in ["ERROR", "FATAL"]
+            e
+            for e in strip_ccr_sentinels(compressed_data["entries"])
+            if e["level"] in ["ERROR", "FATAL"]
         ]
 
         # CRITICAL: 100% of errors must be preserved
@@ -306,7 +309,7 @@ class TestMCPErrorPreservation:
         # Should preserve some messages with error keywords (SmartCrusher detects these)
         error_msgs = [
             m
-            for m in compressed_data["messages"]
+            for m in strip_ccr_sentinels(compressed_data["messages"])
             if any(kw in m["text"].lower() for kw in ["error", "failed", "exception"])
         ]
         assert len(error_msgs) > 0, "Should preserve some error messages"
@@ -327,7 +330,9 @@ class TestMCPErrorPreservation:
 
         compressed_data = json.loads(result.compressed_content)
         compressed_errors = [
-            r for r in compressed_data["rows"] if "error" in str(r["status"]).lower()
+            r
+            for r in strip_ccr_sentinels(compressed_data["rows"])
+            if "error" in str(r["status"]).lower()
         ]
 
         # Should preserve most error rows
@@ -350,7 +355,7 @@ class TestMCPErrorPreservation:
         # Should preserve at least some bugs
         compressed_bugs = [
             i
-            for i in compressed_data["issues"]
+            for i in strip_ccr_sentinels(compressed_data["issues"])
             if any(label in ["bug", "critical", "urgent", "blocker"] for label in i["labels"])
         ]