booking-notify

Warn

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEPERSISTENCEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The documentation in webhook-relay.md instructs the user to include the Cal.com API key as a query parameter in a curl request: curl -s -X POST "https://api.cal.com/v1/webhooks?apiKey=$CALCOM_API_KEY". This practice exposes sensitive credentials in shell history, process listings, and potentially in network/server logs. Additionally, secrets are passed to Cloud Run as environment variables via --set-env-vars, which is less secure than using dedicated secret management services.
  • [PERSISTENCE]: The skill implements a persistence mechanism described in sync-config.md and SKILL.md by using launchd to schedule a synchronization script to run automatically every 6 hours. While intended for its core functionality, this allows code to persist and execute across user sessions.
  • [EXTERNAL_DOWNLOADS]: The SKILL.md file contains a build step that executes bun install within the calcom-cli directory. This command fetches third-party dependencies from external registries, introducing a supply-chain risk.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process booking data from the external Cal.com API, which includes attendee names and notes that could contain malicious instructions.
  • Ingestion points: Booking payloads fetched from the Cal.com API as described in sync-config.md.
  • Boundary markers: Notification templates in notification-templates.md use placeholder interpolation (e.g., {{notes}}), but there are no instructions for the agent to ignore or neutralize potential commands embedded within that data.
  • Capability inventory: The skill has the capability to execute shell commands (Bash) and make network requests (curl) based on the processing of this data.
  • Sanitization: Although notification-templates.md mentions a stripHtmlForPushover() function, no sanitization logic for prompt injection or executable content is visible in the provided files.
  • [COMMAND_EXECUTION]: The skill performs various automated shell operations including building binaries (bun run build) and deploying infrastructure to Google Cloud Platform (gcloud run deploy), which involve high-privilege interactions with the user's local environment and cloud accounts.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 7, 2026, 02:12 AM