The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the apify CLI and jq to perform data retrieval and processing. It instructs the agent to execute shell commands such as apify actors call and apify datasets get-items to interact with the Apify platform. Output is redirected to temporary files (e.g., /tmp/results.json) and then processed using jq.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the apify-cli package via npm (npm i -g apify-cli). While this is a tool provided by the vendor, it represents an external dependency required for the skill to function.
[PROMPT_INJECTION]: The skill ingests and processes untrusted data from various external web sources (Google search results, Reddit, Twitter/X, NVD, CISA catalog, and arbitrary website content). This creates an attack surface for indirect prompt injection where malicious instructions embedded in these external sources could attempt to influence the agent's behavior.
Ingestion points: Data retrieved from actors like apify/google-search-scraper, trudax/reddit-scraper, and apify/website-content-crawler in SKILL.md.
Boundary markers: The skill instructions do not specify the use of delimiters or clear "ignore embedded instructions" warnings for the agent when processing external data.
Capability inventory: The agent has the capability to execute shell commands (apify, jq) and write to the local filesystem (/tmp/results.json).
Sanitization: There is no explicit sanitization or filtering of the external content described before the agent analyzes it or passes it to processing tools.

apify-osint-threat-intel