The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute the icp CLI tool with various subcommands including identity link, deploy, and canister settings. These operations interact with the local filesystem and the network.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted inputs from the user, specifically the subnet-id and console-origin, which are then interpolated into shell commands.
Ingestion points: User-provided strings for engine configuration in SKILL.md (Step 1 and Step 3).
Boundary markers: Absent. There are no instructions for the agent to sanitize or validate these inputs before including them in shell command strings.
Capability inventory: The skill performs subprocess execution (icp commands), file reads (icp.yaml, package.json), and network requests (via the CLI).
Sanitization: Absent. The agent is not instructed to escape shell characters or validate the format of the provided IDs or URLs.
[DYNAMIC_EXECUTION]: The metadata configuration examples for icp.yaml utilize shell command substitution $(...) (e.g., $(git rev-parse HEAD), $(node -p ...)). The skill explicitly notes that these values are interpolated into shell commands at build time, creating a vector for executing arbitrary code defined within the project's configuration files.
[EXTERNAL_DOWNLOADS]: The skill references external domains for documentation (cli.internetcomputer.org) and default service endpoints (opencloud.org). These references are consistent with the infrastructure of the author (dfinity).

deploy-to-cloud-engine