The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill's error handling instructions explicitly direct the agent to retry commands with elevated permissions (e.g., sudo) if it encounters sandbox or permission failures during package installation.
[INDIRECT_PROMPT_INJECTION]: The skill consumes untrusted data from Snyk security scans and uses this information to determine which code to modify and which shell commands to execute.
Ingestion points: Security scan results provided by the mcp_snyk_snyk_code_scan and mcp_snyk_snyk_sca_scan tools (SKILL.md).
Boundary markers: None. The instructions lack explicit delimiters or safety markers to differentiate between scan data and instructions.
Capability inventory: The skill utilizes Write, Edit, and Bash tools to modify project files, resolve dependencies, and manage Git repositories (SKILL.md).
Sanitization: The skill does not specify any validation or sanitization logic for the content of the scan results before they are interpolated into remediation workflows.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform a wide range of development actions, including using git, gh (GitHub CLI), and various package managers like npm, yarn, pip, mvn, and go to apply fixes and create pull requests.
[EXTERNAL_DOWNLOADS]: The remediation workflow involves downloading and installing package upgrades from remote registries via standard package managers to resolve identified vulnerabilities.

snyk-fix