The Agent Skills Directory

Automated Validation via Command Execution: The skill utilizes a linter checklist (assets/linter-checklist.md) that executes shell commands and Python scripts. While running dynamic code is a security consideration, in this context it is used for mechanical verification of the specification, such as ensuring character encoding and schema compliance.\n- Remote Tooling for Diagram Validation: One of the linter checks uses npx to run the @mermaid-js/mermaid-cli tool for diagram verification. Downloading and executing tools from external registries is a standard development pattern, and here it involves a well-known service from a reputable source.\n- Sanitization of Untrusted PR Content: The skill processes external data from pull requests, which is a potential surface for indirect prompt injection. The skill mitigates this through several hardening techniques:\n
Structural Validation: It enforces strict JSON schema validation for all subagent outputs to ensure data integrity.\n
Encoding Enforcement: It mandates that all written output be within the printable ASCII range, which protects against encoding-based attacks or hidden characters.\n
Architectural Constraints: It employs a single-writer model where only the orchestrator interacts with external APIs, limiting the capabilities of individual subagents.

apm-spec-guardian