The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill serves as a reference for managing server resources, which involves processing potentially untrusted data from the managed system (e.g., logs or database query results). This creates a surface for indirect prompt injection.
Ingestion points: The skill facilitates interaction with tools that ingest external data, such as SiteLogs, Grep, MysqlQuery, and BashStatus (found in SKILL.md).
Boundary markers: The skill instructions (e.g., in the Workflow and Rules sections of SKILL.md) mandate the use of desensitized fragments and forbid copying full credentials into outputs.
Capability inventory: The skill catalogs a toolset with significant capabilities, including Bash for shell execution, Write for file modification, and various project deletion tools.
Sanitization: The Rules section explicitly forbids the collection or storage of user credentials and requires TLS validation for all external network operations.
[COMMAND_EXECUTION]: The skill documents tools with high-risk capabilities, specifically the Bash tool which allows for arbitrary shell command execution. This is an intended primary function for the documented server administration use case. The skill provides safety guardrails by categorizing these tools as 'high' risk and requiring a structured mcp-tool-plan and explicit user authorization before any high-risk action is taken.
[DATA_EXFILTRATION]: The skill provides guidance for tools that can access sensitive system information, such as Read, DatabaseBackup, and SSHInfo. The instructions include specific safety rules to prevent the exfiltration of credentials, requiring that all external calls use trusted endpoints and maintaining strict TLS verification to protect data in transit.

bt-mcp-tools