From ec4ac6ef8e35d7c7469051e234fe002693c1e625 Mon Sep 17 00:00:00 2001
From: MrSphay <leifwilkens@gmail.com>
Date: Wed, 8 Jul 2026 13:45:17 +0200
Subject: [PATCH] Decode git forge content responses for agents

---
 src/integrations.py                           | 61 +++++++++++++++++++
 .../test_integrations_api_call_truncation.py  | 33 +++++++++-
 2 files changed, 91 insertions(+), 3 deletions(-)

diff --git a/src/integrations.py b/src/integrations.py
index 92b628f..a94e09b 100644
--- a/src/integrations.py
+++ b/src/integrations.py
@@ -1,3 +1,4 @@
+import base64
 import json
 import os
 import uuid
@@ -372,6 +373,61 @@ def _find_integration(identifier: str) -> Optional[Dict[str, Any]]:
     return None
 
 
+def _decode_repository_content_response(
+    data: Any,
+    *,
+    preset: str,
+    path: str,
+) -> Any:
+    """Make git forge contents responses usable for agents.
+
+    Gitea/GitHub contents endpoints return file bytes as a Base64 string.
+    Sending that raw blob to the model wastes context and encourages the agent
+    to manually decode it in chat. Decode text files here and omit the raw blob
+    from the tool output.
+    """
+
+    if not isinstance(data, dict):
+        return data
+
+    if preset not in {"gitea", "github"} and "/contents/" not in path:
+        return data
+
+    if data.get("encoding") != "base64" or not isinstance(data.get("content"), str):
+        return data
+
+    normalized = dict(data)
+    raw_content = normalized.pop("content", "")
+    try:
+        decoded_bytes = base64.b64decode(raw_content, validate=False)
+    except Exception:
+        normalized["content_omitted"] = True
+        normalized["content_error"] = "Base64 content could not be decoded"
+        return normalized
+
+    normalized["content_omitted"] = True
+    normalized["content_original_encoding"] = "base64"
+    normalized["decoded_size_bytes"] = len(decoded_bytes)
+
+    try:
+        decoded_text = decoded_bytes.decode("utf-8")
+    except UnicodeDecodeError:
+        normalized["decoded_content"] = None
+        normalized["content_error"] = "Decoded content is not valid UTF-8"
+        return normalized
+
+    decoded_limit = 8000
+    if len(decoded_text) > decoded_limit:
+        normalized["decoded_content"] = decoded_text[:decoded_limit]
+        normalized["decoded_truncated"] = True
+        normalized["decoded_total_chars"] = len(decoded_text)
+    else:
+        normalized["decoded_content"] = decoded_text
+        normalized["decoded_truncated"] = False
+
+    return normalized
+
+
 async def execute_api_call(
     integration_id: str,
     method: str,
@@ -490,6 +546,11 @@ async def execute_api_call(
         if "application/json" in content_type:
             try:
                 data = response.json()
+                data = _decode_repository_content_response(
+                    data,
+                    preset=preset,
+                    path=path,
+                )
                 full = json.dumps(data, indent=2, ensure_ascii=False)
                 if len(full) > 12000:
                     if isinstance(data, list):
diff --git a/tests/test_integrations_api_call_truncation.py b/tests/test_integrations_api_call_truncation.py
index bf1ec7d..5cd51d5 100644
--- a/tests/test_integrations_api_call_truncation.py
+++ b/tests/test_integrations_api_call_truncation.py
@@ -4,6 +4,7 @@ Covers:
   (a) Large JSON list response -> sentinel appended, valid JSON returned
   (b) Small response -> returned unchanged, no truncation
 """
+import base64
 import json
 import sys
 import os
@@ -72,7 +73,7 @@ def _make_response(json_data, status=200):
     return resp
 
 
-async def _call(json_data, status=200):
+async def _call(json_data, status=200, *, integration=None, path="/items"):
     mock_resp = _make_response(json_data, status)
 
     mock_client = AsyncMock()
@@ -81,13 +82,13 @@ async def _call(json_data, status=200):
     mock_client.request = AsyncMock(return_value=mock_resp)
 
     with (
-        patch.object(integrations, "_find_integration", return_value=DUMMY_INTEGRATION),
+        patch.object(integrations, "_find_integration", return_value=integration or DUMMY_INTEGRATION),
         patch("httpx.AsyncClient", return_value=mock_client),
         # api.example.com doesn't resolve; the SSRF guard would fail closed.
         # These tests are about truncation, so stub the guard open.
         patch("src.url_safety.check_outbound_url", return_value=(True, "ok")),
     ):
-        return await integrations.execute_api_call("test_integ", "GET", "/items")
+        return await integrations.execute_api_call("test_integ", "GET", path)
 
 
 async def _call_with_integration(integration, path="/items"):
@@ -230,6 +231,32 @@ async def test_small_json_dict_not_truncated():
     assert "_truncated" not in parsed
 
 
+@pytest.mark.asyncio
+async def test_gitea_contents_response_decodes_base64_and_omits_raw_blob():
+    text = 'APP_VERSION = "1.0.1"\n'
+    payload = {
+        "name": "constants.py",
+        "path": "src/constants.py",
+        "encoding": "base64",
+        "content": base64.b64encode(text.encode("utf-8")).decode("ascii"),
+    }
+    integration = {**DUMMY_INTEGRATION, "preset": "gitea"}
+
+    result = await _call(
+        payload,
+        integration=integration,
+        path="/api/v1/repos/MrSphay/odysseus/contents/src/constants.py",
+    )
+
+    assert result.get("exit_code") == 0
+    body = result["output"].split(chr(10), 1)[1]
+    parsed = json.loads(body)
+    assert parsed["decoded_content"] == text
+    assert parsed["decoded_truncated"] is False
+    assert parsed["content_omitted"] is True
+    assert "content" not in parsed
+
+
 @pytest.mark.asyncio
 async def test_list_truncation_respects_limit_including_sentinel():
     """After list truncation the total serialized body must not exceed 12000 chars,