The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill fetches a "local-triage playbook" from https://www.react.doctor/prompts/react-doctor-agent.md and explicitly instructs the agent to "follow every step in it". This creates a remote instruction injection vector where a third-party host can dynamically control agent logic and behavior without a skill update.
[EXTERNAL_DOWNLOADS]: The skill frequently executes npx react-doctor@latest, which downloads and runs the latest version of a package from the public npm registry at runtime. This practice is susceptible to supply chain attacks and lacks version pinning for security.
[DATA_EXFILTRATION]: The performance scanning capability (npx react-doctor@latest scan <url>) opens browser profiles and records DevTools traces, which are identified as "sensitive local application data". The skill also suggests using --cdp <remote-debugging-url> to access authenticated browser states, creating a path for potential exposure of sensitive session data or credentials.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted markdown data from remote URLs, including the main playbook and per-rule prompts at https://www.react.doctor/prompts/rules/<plugin>/<rule>.md. While the skill contains a warning to treat this as untrusted input, the instruction to "follow every step" in the remote content provides a massive surface for instructions to bypass local safety guidelines.
[COMMAND_EXECUTION]: The skill executes multiple shell commands using npx and curl, and performs local file modifications on package.json and doctor.config.* files based on tool outputs and remote instructions.

react-doctor