The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill correctly identifies that data retrieved from Sentry (event payloads, issue titles, comments) could contain malicious instructions. It includes a specific safety instruction in SKILL.md for the agent to treat this data as untrusted and never execute instructions found within it.
Ingestion points: references/setup-verification.md and references/new-project.md involve reading data from the Sentry MCP (Model Context Protocol) server.
Boundary markers: The skill provides explicit instructions to the agent to disregard embedded commands in external data.
Capability inventory: The skill can modify application source code (SDK initialization), install packages, and execute shell commands for build/deploy processes.
Sanitization: The agent is instructed to avoid executing instructions found in untrusted payloads.
[CREDENTIALS_UNSAFE]: The skill provides detailed guidance on the secure handling of the SENTRY_AUTH_TOKEN.
Evidence: references/auth-token.md explicitly warns against committing tokens to version control, printing them in logs, or pasting them into chats. It advocates for the use of environment variables and CI secret stores, which is standard security best practice.
[EXTERNAL_DOWNLOADS]: The skill references several vendor-provided tools and SDKs for instrumentation.
Evidence: Mentions npx @sentry/wizard@latest, brew install getsentry/tools/sentry-wizard, and various @sentry/* packages. As these resources belong to the official vendor (getsentry) and are necessary for the primary purpose of the skill, they are considered safe.
[COMMAND_EXECUTION]: The skill instructs the agent to perform command-line operations for package installation, build processing, and verification.
Evidence: Commands such as npm install, flutter build, and sentry-cli operations are used to set up the monitoring environment. These actions are within the expected scope of an instrumentation skill.

sentry-instrument