Security audit
Security checks for vulnerabilities and agentic risk
This skill is a narrow Google Sheets helper, but it uses a hardcoded service-account credential path and can modify a live family school schedule without a strong confirmation step.
Only install this if you own or are authorized to manage the referenced school-run spreadsheet. Before use, make sure the service account is scoped only to that spreadsheet, and require the agent to show the exact sheet tab, range, old value, and new value before any update.
The skill explicitly supports direct updates to a live spreadsheet but does not require explicit user confirmation, warn that writes affect real data, or constrain when modification is appropriate. In practice, an agent could perform unintended or irreversible edits to a family schedule based on ambiguous prompts or mistaken row selection.
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
- Always check current data before appending or updating.
- Use `gws sheets spreadsheets values get` to inspect specific dates.
- Use `gws sheets spreadsheets values update` to modify existing rows.
- **Dry-run first** for any write operation.
## Command Templates
The skill documentation embeds and depends on a specific local service-account credential file path, revealing sensitive environmental details and normalizing direct use of privileged credentials from the host. In an agent setting, this can encourage unauthorized access patterns, credential misuse, or lateral discovery of secrets beyond the narrow school-run scheduling purpose.
The sheet schema hard-codes an English-language field description ('Who is responsible for dropping Damian and Zachary off at school') and the entire skill is written as English-only operational guidance, with no mention of language choice or locale rationale. Under SQP-3, forcing a specific language without opt-in can be a natural-language policy concern.
No suspicious patterns detected.