overleaf-sync
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows credential management best practices by explicitly prohibiting the storage or transmission of authentication tokens within chat history, environment variables, or disk files. It leverages OS-level credential helpers (macOS Keychain, Windows Credential Manager) to handle authentication securely.
- [SAFE]: The skill implements automated technical controls to prevent data leakage, including a pre-commit hook that scans for Overleaf token patterns (
olp_[A-Za-z0-9]{20,}) and aborts commits if they are detected. - [SAFE]: The setup process is designed to be performed by the user in a local terminal rather than through the agent. The setup script includes a safeguard that refuses to execute in non-TTY environments (like an agent's shell harness) to ensure human-in-the-loop interaction for sensitive configuration.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data pulled from Overleaf projects which could contain instructions from collaborators. This risk is effectively mitigated by the skill's mandatory "Diff protocol," which requires the agent to perform granular reviews of changes (checking for half-finished sentences, numerical claim changes, etc.) and seek user approval before syncing prose changes or structural updates.
- Ingestion points:
git pullfrom the Overleaf project repository intopaper-overleaf/. - Boundary markers: None explicitly defined in the raw files, but the skill enforces a behavioral boundary via diff verification before merging into the local
paper/directory. - Capability inventory: Uses
Bashfor git and rsync operations, andEdit/Writefor file modifications. - Sanitization: Relies on a rigorous diff protocol and selective syncing using
rsyncor theEdittool to filter out suspicious or malformed content before it enters the primary working directory.
Audit Metadata