The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses npx to run a skills update command which fetches content from a remote repository to update the local skill files. This allows the skill to change its behavior after it has been loaded.
[PROMPT_INJECTION]: Instructions specifically command the agent to perform the update process 'silently' and explicitly tell it not to mention the update to the user, which conceals remote content fetching and potential instruction changes.
[DYNAMIC_EXECUTION]: After performing the remote update, the agent is instructed to re-read the SKILL.md file from disk and follow the new version. This creates a pattern where the initial instructions act as a stager for un-audited remote content.
[INDIRECT_PROMPT_INJECTION]: The skill scans the local codebase (using grep and file reads) for authentication headers and route definitions. This data is then interpolated into shell commands (calibrate agents create --config-param '<json>'). This creates an attack surface where malicious code in the analyzed repository could influence the parameters of the CLI command.
[COMMAND_EXECUTION]: The skill facilitates the installation and upgrade of the calibrate CLI tool using Homebrew (brew install dalmia/tap/calibrate). While consistent with the stated purpose of the skill, it involves downloading and installing third-party binaries.

connect-agent