The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/install.sh script downloads the Lightpanda browser binary from its official GitHub repository (lightpanda-io/browser). The script implements an integrity check by comparing the downloaded file's SHA256 hash against a checksum retrieved from the GitHub API.
[COMMAND_EXECUTION]: The skill requires the execution of the lightpanda binary on the host system. The installation script uses chmod a+x to grant execution permissions to the downloaded binary.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it retrieves and processes untrusted web content which could contain malicious instructions.
Ingestion points: Web content is ingested through the goto tool and presented to the agent as markdown or a semantic tree (SKILL.md).
Boundary markers: The skill instructions do not specify the use of delimiters or 'ignore' instructions to isolate web content from the agent's main objective.
Capability inventory: The agent has the ability to interact with web content using tools like click, fill, and evaluate (JavaScript execution), which could be misused if influenced by injected instructions.
Sanitization: There is no evidence of sanitization or filtering of the retrieved web content before it is processed by the agent.

Lightpanda