skills/matematicsolutions/awesome-matematic-skills-pl/matematic-workspace-backup/Gen Agent Trust Hub
matematic-workspace-backup
Warn
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to download the
gogcliv0.14.0 binary from an external GitHub repository (github.com/steipete/gogcli). While it recommends verifying the SHA256 checksum, this pattern involves executing code from a non-standard third-party source. - [REMOTE_CODE_EXECUTION]: The workflow involves executing the downloaded
gogclibinary directly in the environment. Additionally, Step 6 references the execution of a shell script./build-safe.shto compile safety profiles, which constitutes dynamic execution of local scripts. - [COMMAND_EXECUTION]: The skill makes extensive use of the
Bashtool to perform system-level operations, including automated backups (gog backup push), file verification, and cleaning up plaintext data usingRemove-Item -Recurse -Force. It also configures persistence via Windows Task Scheduler and Cron for daily and weekly execution. - [DATA_EXFILTRATION]: The primary purpose of the skill is to read data from Google Workspace (Gmail, Drive) and push it to an external private Git repository. This involves the systematic egress of potentially sensitive data to a third-party storage location managed via OAuth and Service Account credentials.
- [PROMPT_INJECTION]: The skill processes untrusted content from Google Workspace (emails and files). Although it includes
pii-egress: maskedand suggests bounded queries, the ingestion of external data that is later processed by an agent creates a surface for indirect prompt injection. - Ingestion points: Gmail bodies/metadata and Google Drive files are read during the backup process (Step 4, Step 5).
- Boundary markers: The skill suggests using
--queryand--maxto limit the scope of first backups, but no specific prompt delimiters are mentioned for preventing the agent from obeying instructions embedded in emails. - Capability inventory: The skill has access to
Bash,Write, andReadtools, enabling it to write to the file system, execute binaries, and perform network operations. - Sanitization: The metadata manifest is limited to cleartext metadata-only, and the skill frontmatter specifies
pii-egress: maskedto reduce data exposure.
Audit Metadata