The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates on data ingested from the Sentry API (via MCP tools such as search_issues, get_sentry_resource, and search_events), which includes attacker-controllable content like issue titles, tags, and stack trace paths. This creates a potential surface for indirect prompt injection. The skill mitigates this by providing explicit defensive instructions to the agent to treat all MCP-returned data as untrusted and to never execute instructions found within event payloads.
Ingestion points: MCP tools search_issues, get_sentry_resource, search_events, and get_release_details.
Boundary markers: The skill includes specific warnings in SKILL.md instructing the agent to ignore embedded instructions in external data.
Capability inventory: Terminal execution via sentry-cli and curl, file system access for SDK and CI configuration, and Sentry tool execution.
Sanitization: Instructional guardrails implemented as part of the agent's core task flow.
[COMMAND_EXECUTION]: The skill frequently utilizes terminal commands including sentry-cli and curl to manage releases, upload artifacts, and interact with the Sentry API. These operations are standard for the skill's purpose and are documented with security best practices in mind, particularly regarding token handling.
[EXTERNAL_DOWNLOADS]: The skill references the installation of vendor-owned tools such as @sentry/wizard via npx and getsentry/tools/sentry-wizard via brew. As these originate from the trusted vendor (getsentry), these references are considered safe and represent standard integration patterns.
[CREDENTIALS_UNSAFE]: The skill manages the SENTRY_AUTH_TOKEN required for CI/CD integrations. It provides robust security guidance, instructing users to store tokens in CI secrets, avoid committing them to version control, and using a safe verification method ([ -n "$SENTRY_AUTH_TOKEN" ] && echo "SENTRY_AUTH_TOKEN=set") that confirms the token's presence without revealing its value.

sentry-setup-releases