The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the official CLI installation script from the vendor's GitHub repository.
Evidence: curl -fsSL https://raw.githubusercontent.com/InsForge/insta-cli/main/install.sh | sh in SKILL.md. This is documented as the standard installation method for the InsForge toolset.
[REMOTE_CODE_EXECUTION]: The installation process involves executing a remote script via shell.
Evidence: The skill instructs the agent to install the CLI tool using a piped shell command when it is not present in the environment. This execution is limited to official vendor resources.
[COMMAND_EXECUTION]: The skill relies on the execution of the insta CLI to manage cloud resources.
Evidence: The instructions utilize numerous subcommands (e.g., service add, deploy, secrets bind) to interact with the InstaCloud platform. These commands are executed within a governed agent session.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests project configuration and environment data.
Ingestion points: The agent reads local configuration files such as .insta/project.json and environment variables from .env.
Boundary markers: The skill encourages the use of the --json flag to ensure structured and predictable parsing of tool outputs.
Capability inventory: The agent can execute shell commands, perform network requests to the InstaCloud API, and modify cloud infrastructure.
Sanitization: The skill implements a mandatory 'agent policy' system that enforces restrictions and requires human-in-the-loop approvals for sensitive mutations and credential access.

insta