The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from the Sentry MCP (including event payloads, issue titles, and comments) and reads project source files. When combined with capabilities such as package installation (pip, npm, composer, etc.), file modification, and shell command execution, this creates a surface for indirect prompt injection. The skill includes an explicit instruction to the agent to treat MCP data as untrusted and to avoid executing instructions found within it, which serves as a significant policy-based mitigation.
[DATA_EXFILTRATION]: Automated scanners flagged the URL https://analytics.third-party.com as related to a Botnet. Analysis of the source file references/sdks/svelte/session-replay.md reveals that this URL is used exclusively as a placeholder example in a configuration snippet for the networkDetailDenyUrls option, which is designed to exclude third-party analytics from data collection. The usage is defensive and pedagogical.
[EXTERNAL_DOWNLOADS]: Automated scanners flagged the file references/sdks/svelte/session-replay.md as malware (FileRepMalware [Misc]). This file is a Markdown documentation resource authored by Sentry. Manual review indicates the detection is a false positive likely triggered by the technical code examples or the presence of the placeholder analytics domain. The file contains only legitimate instrumentation guidance and no executable or malicious logic.

sentry-instrument