The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches infrastructure metadata, API documentation, and provisioning data from its official domain, cohesivity.ai. These requests include standard telemetry such as User-Agent and request IP for service-side abuse prevention.
[COMMAND_EXECUTION]: Provides a bash script for initializing the .cohesivity credential file. The implementation follows security best practices by using umask 077 to ensure restrictive permissions, validating the environment using git check-ignore, and protecting against symlink and race-condition attacks during file creation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process live API documentation from remote endpoints. It includes proactive security instructions to prevent the agent from following directives embedded in that external content, requiring it to treat documentation solely as reference material.
Ingestion points: Remote documentation from https://cohesivity.ai/offerings/<name> and https://cohesivity.ai/llms.txt.
Boundary markers: Explicit instructions in the 'Best Practices' and 'Security & Safety Notes' sections to treat live docs as reference and not instructions.
Capability inventory: Shell command execution for tenant initialization, local credential file management, and resource provisioning via network requests.
Sanitization: Mandatory requirement to ignore embedded directives in fetched documents and use non-default User-Agents for identification.

cohesivity