The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and follow integration instructions from external plugin metadata, which could contain malicious prompts designed to hijack the agent's behavior.
Ingestion points: The skill reads the usagePrompt and readme fields from the output of the orderly-devkit view <pluginId> command in Step 2 of SKILL.md.
Boundary markers: Absent. The instructions from the usagePrompt are treated as authoritative guidance without delimiters or warnings to ignore embedded instructions.
Capability inventory: The agent has the capability to modify project source code (React components), update package.json manifests, and execute shell commands like pnpm install.
Sanitization: Absent. There is no evidence of sanitization or validation performed on the external metadata before it is processed by the agent.
[COMMAND_EXECUTION]: The skill utilizes shell commands to interact with the project environment and external developer tools.
Evidence: The skill uses rg (ripgrep) to search for components, npx orderly-devkit to view plugin metadata, and pnpm install to manage dependencies. While these are standard development tasks, the use of user-provided plugin IDs in shell commands without explicit sanitization presents a potential surface for shell argument injection.

orderly-plugin-add