The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute 'npx -y modern-web-guidance@latest' to retrieve guidelines. This pattern represents the execution of arbitrary remote code from an unpinned and untrusted external package.
[EXTERNAL_DOWNLOADS]: The skill downloads external content from the npm registry via the 'modern-web-guidance' package without source verification.
[COMMAND_EXECUTION]: The skill utilizes several high-risk command patterns:
Managing the local git repository (diff, add, commit) to implement code changes.
Executing project-specific development and build scripts (e.g., 'npm run dev').
Running browser automation tests via 'npx playwright test'.
Dynamically creating and executing a standalone test harness in the system's temporary directory.
[DATA_EXFILTRATION]: The skill provides templates and instructions for browser automation (Playwright) that can access the user's active browser session and authenticated requests. The 'Origin-replay pattern' described in the healing phase provides a mechanism for intercepting and re-issuing authenticated requests, which could be exploited to access or exfiltrate sensitive user data or perform unauthorized actions in the user's session context.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data from external guides and the target codebase to drive its integration logic.
Ingestion points: Project source code and external guidelines fetched via 'npx'.
Boundary markers: Absent; the skill does not explicitly instruct the agent to use safety delimiters for these specific inputs during its internal reasoning.
Capability inventory: Extensive filesystem access, shell command execution, git manipulation, and authenticated browser interaction.
Sanitization: While the skill provides detailed guidance for sanitizing the output tools it creates for the web app, it lacks internal sanitization for the inputs it processes from the user's project or remote sources.

webmcpify