The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface where the agent processes untrusted data retrieved from the Sentry API. It mitigates this risk by providing mandatory safety instructions in SKILL.md to treat all MCP-returned data as untrusted and specifically warns against executing any instructions found within event payloads or issue titles. Evidence chain:
Ingestion points: search_issues, get_sentry_resource, and get_release_details (MCP tools).
Boundary markers: Explicit instructions in SKILL.md to ignore embedded instructions in external data.
Capability inventory: Shell command execution via sentry-cli and npx tools.
Sanitization: Prompt-based directives for the agent to validate and scope the processing of retrieved data.
[DATA_EXFILTRATION]: The skill requires the use of SENTRY_AUTH_TOKEN for interacting with the Sentry API. It adheres to security best practices by guiding the user to store tokens in secure environments (CI secrets, environment variables) and provides methods to verify their presence without revealing the actual values. All network operations are directed to the official Sentry domain (sentry.io), which is the authoritative endpoint for the skill's purpose.
[COMMAND_EXECUTION]: The instructions involve executing standard Sentry CLI and development tools such as sentry-cli, npx @sentry/wizard, and brew. These commands are used transparently for their intended purposes of project configuration, artifact upload, and setup verification, with no evidence of hidden or malicious execution patterns.
[EXTERNAL_DOWNLOADS]: The skill references the installation of various official Sentry SDKs and management tools. These resources originate from trusted package registries (NPM) or official repositories owned by the vendor (getsentry), presenting no risk to the development environment when used as instructed.

sentry-setup-releases