The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads and executes the @reticlehq/server package from the NPM registry using npx. This is a vendor-provided resource necessary for the skill's functionality.
[DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically read shell scripts from the local package.json file and execute them in the background to start a development server.
[COMMAND_EXECUTION]: It uses several shell commands including npx and background process management. It specifically instructs the agent to use autonomous modes or flags (like npx -y) to avoid pausing for human interaction or approval.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources which could contain malicious instructions.
Ingestion points: Reads package.json content and monitors web server response bodies (via bodyContains and until assertions).
Boundary markers: None explicitly defined in the instructions for handling these data sources.
Capability inventory: Includes the ability to execute shell commands (npx), start background processes, and call MCP tools for browser control (reticle_lease).
Sanitization: No specific sanitization or validation of the ingested scripts or response bodies is described.
[DATA_EXPOSURE]: The skill accesses and stores state information in ~/.reticle. While this is outside the project workspace, the skill explicitly discloses this behavior to the user.

verify-unattended