The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from GitHub Dependabot alerts which could potentially contain malicious instructions. The skill proactively mitigates this risk by instructing the agent to treat all API responses and user-provided URLs as data only and to ignore any commands embedded within alert descriptions or metadata.
[COMMAND_EXECUTION]: The skill uses tools like gh api, git, and yarn to perform its tasks. These operations are scoped to the specific repository and are preceded by explicit instructions to wait for user review and approval before making significant changes like committing code.
[DATA_EXFILTRATION]: All network and file operations are directed at the official GitHub repository for the project (getsentry/sentry-javascript). No sensitive data is transmitted to unauthorized third-party services.

fix-security-vulnerability