The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill relies on an external binary and browser extension called OpenCLI that can execute arbitrary code in the browser context with user privileges.
[DYNAMIC_EXECUTION]: Multiple core scripts such as scripts/safe-fill.mjs and scripts/ground-truth.mjs construct and execute JavaScript strings within the browser session via opencli ... eval. These strings often incorporate unvalidated data extracted from external websites.
[INDIRECT_PROMPT_INJECTION]: The skill crawls and processes untrusted content from blog comments and public directories (ingestion points in scripts/harvest-commenters.mjs and scripts/discovery-queue.mjs). It lacks explicit boundary markers or sanitization to prevent the agent from executing instructions embedded in this external data.
[EXTERNAL_DOWNLOADS]: The skill requires users to install a third-party binary and extension from github.com/yan-labs/OpenCLI, a vendor repository, which is necessary for all browser-based tasks.
[DATA_EXFILTRATION]: Automated scans confirmed the presence of a malicious URL https://submitaitools.org/ in the data/paid-platforms.json file. Additionally, the file itself was flagged as malware by security scanners, indicating it may contain malicious data or script patterns.

backlink