The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external shell commands such as agent-browser and npx @claude-flow/cli to perform its operations.
[EXTERNAL_DOWNLOADS]: Uses npx to download and execute the @claude-flow/cli package from the NPM registry at runtime.
[REMOTE_CODE_EXECUTION]: Running npx allows for the execution of remotely hosted code packages during skill execution.
[DATA_EXFILTRATION]: The skill manages authentication states by saving and loading auth.json files, which typically contain sensitive session tokens or cookies. Additionally, it can capture and return screenshots and page content to the agent.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests untrusted data from external websites via snapshots and text extraction, which could contain malicious instructions designed to manipulate the agent's behavior.
Ingestion points: Web content processed via agent-browser snapshot, get text, and get html (SKILL.md).
Boundary markers: None explicitly defined to separate untrusted web content from system instructions.
Capability inventory: Subprocess calls for browser interaction, file-write for auth state, and network access via the browser (SKILL.md).
Sanitization: No explicit sanitization or filtering of scraped content is mentioned.

browser