[BUG] GitHub Integration Cannot Properly Read File Contents – Regression in Project Knowledge

Environment

• Platform (select one):
• [x] Claude Pro (GitHub Integration)
• [ ] Anthropic API
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other:
• Claude CLI version: N/A (Web UI)
• Operating System: Windows 11
• Terminal: N/A

---

Bug Description

GitHub integration within Project Knowledge is broken. Repositories and file lists are visible, but file contents are not read correctly. The system returns only partial or corrupted code fragments, which makes code analysis unreliable and breaks developer workflows.

---

Steps to Reproduce

1. Connect any GitHub repository to Project Knowledge.
2. Request a specific file from the integration without modifying it.

• Observe that instead of returning the original file content, the system generates a completely new version of the file.

1. Alternatively, ask the system to add a single function to an existing file.

• Instead of just adding the function, you will see that many other changes appear, including removals and unrelated modifications.

---

Expected Behavior

Full file content should be fetched and displayed correctly for accurate code analysis.

---

Actual Behavior

Only fragments or incorrect versions of files are returned. In some cases, the system appears to generate a new file structure instead of reading the actual file, leading to significant discrepancies.

---

Additional Context

Screenshots attached show a comparison between the file fetched from the GitHub integration and the same file uploaded directly.
🔴 The file from GitHub appears regenerated and incomplete
🟢 The file uploaded manually shows the correct code structure and includes all requested elements

I already disconnected and reconnected GitHub, followed all suggested support steps — no effect. Support confirmed this is an ongoing sync incident with no ETA.

This issue has persisted for over a week and requires immediate engineering attention.

!Image
!Image