The Agent Skills Directory

[DATA_EXPOSURE_&_EXFILTRATION]: The skill provides comprehensive guidance on managing sensitive authentication tokens (SENTRY_AUTH_TOKEN). It explicitly instructs the user and the agent never to commit these tokens to version control, never to inline them in public configuration files, and never to print them in logs. It recommends using environment variables and secret management systems provided by CI/CD platforms.
[UNVERIFIABLE_DEPENDENCIES_&_REMOTE_CODE_EXECUTION]: The skill references several external packages and scripts for installation (e.g., npm packages like @sentry/react, Python packages like sentry-sdk, and the Sentry Wizard). These resources are all official Sentry products hosted on well-known, trusted domains and registries (npmjs.com, pypi.org, GitHub repositories under the getsentry organization).
[INDIRECT_PROMPT_INJECTION]: The skill identifies that data returned from the Sentry Model Context Protocol (MCP) server, such as event tags, messages, and issue titles, should be treated as untrusted input. It explicitly warns the agent never to execute instructions found inside an event payload or issue title, which is a key mitigation against indirect prompt injection.
[PROMPT_INJECTION]: A deterministic detector flagged a potential concealment issue in the verification steps. However, manual review confirms this is a false positive. The instructions actually advocate for transparency, requiring the agent to show the user exactly what was captured and requesting permission before performing 'cleanup' actions like resolving test issues.

sentry-setup-releases