github/AutoGPT
1
1
Fork 0
mirror of https://github.com/Significant-Gravitas/AutoGPT.git synced 2026-02-06 04:45:10 -05:00
Code Issues Packages Projects Releases Wiki Activity

docs: clarify double-scan behavior is intentional

Browse Source
This commit is contained in:
Nick Tindle
2026-02-05 22:25:00 -06:00
parent cf3c6d0c22
commit 495fa6531d
1 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
docs/platform/workspace-media-architecture.md
View File

@@ -197,13 +197,19 @@ output = await store_media_file(
|-----------|--------|-------|
| `store_media_file()` | ✅ Yes | Scans **all** content before writing to local disk |
| `WorkspaceManager.write_file()` | ✅ Yes | Scans content before persisting (defense in depth) |
| Upload API endpoints | ✅ Yes | Also scan before calling WorkspaceManager (fail fast) |
| `WriteWorkspaceFileTool` | ✅ Yes | Scans before calling WorkspaceManager (fail fast) |
**Defense in depth:** Scanning happens at multiple layers:
1. `store_media_file()` scans everything it downloads/decodes
2. API endpoints scan uploads for early rejection
2. CoPilot tools (e.g., `WriteWorkspaceFileTool`) scan for early rejection
3. `WorkspaceManager.write_file()` scans as a final gate before persistence
**Note on double scanning:** Some paths (like `WriteWorkspaceFileTool`) will scan twice — once at the API/tool layer and once in `WorkspaceManager.write_file()`. This is intentional:
- **First scan (tool layer):** Fail fast, reject bad content before any processing
- **Second scan (persistence layer):** Defense in depth, catches any caller that forgot to scan
The performance cost is acceptable since scanning is fast and security is critical.
### Persistence
| Component | Persists To | Lifecycle |