The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install a CLI tool (belt-sh/cli) and multiple sub-skills (inference-sh/skills) from external repositories. These sources are not verified or listed as trusted providers.
[REMOTE_CODE_EXECUTION]: The skill relies on belt app run to execute logic and applications hosted on the inference.sh platform. This involves running code from a remote, unverified service at runtime.
[COMMAND_EXECUTION]: The skill configuration allows broad execution of commands via allowed-tools: Bash(belt *). This grants the agent the ability to run any sub-command or application available through the belt CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external search results which could contain malicious instructions.
Ingestion points: Data from tavily/search-assistant and exa/search is fetched in SKILL.md via the belt tool.
Boundary markers: There are no delimiters or "ignore embedded instructions" warnings provided to the agent when processing this external content.
Capability inventory: The agent has access to Bash(belt *) which can be used to perform further network operations or execute more remote apps based on injected instructions.
Sanitization: No validation or filtering is applied to the data returned from search tools before it is used to draft the press release.

press-release-writing