The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user or agent to install the falco tool from a personal GitHub repository (github.com/ysugimoto/falco). This involves downloading and executing code from an external, non-official source that is not listed in the trusted vendor list.
Evidence: Installation commands go install github.com/ysugimoto/falco/cmd/falco@latest and git clone https://github.com/ysugimoto/falco.git in SKILL.md and references/understanding-vcl-from-source.md.
[COMMAND_EXECUTION]: The skill relies on the execution of various CLI tools and scripts to perform its functions, including falco, terraform, git, make, and curl for linting, testing, and simulation.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied VCL files and encourages the agent to read their content to understand logic, which presents an attack surface for instructions embedded in external data.
Ingestion points: VCL (.vcl) and XVCL (.xvcl) files, and Terraform plans (planned.out) processed by the falco CLI and read by the agent as instructed in SKILL.md and references/terraform-vcl.md.
Boundary markers: Absent. No instructions are provided to the agent to use delimiters or ignore embedded instructions when reading user-provided VCL files.
Capability inventory: The agent can execute shell commands via falco, write to files (falco fmt -w), and initiate network connections via proxy mode (falco simulate --proxy).
Sanitization: Absent. No mention of sanitizing or validating the content of VCL files before they are processed by the tool or read by the agent.

falco