The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a self-modifying behavior where it executes shell commands to update its own instruction file (SKILL.md) and then requires the agent to reload and follow the new instructions. This allows the skill to bypass the initial static analysis by changing its logic at runtime.
[EXTERNAL_DOWNLOADS]: The skill uses npx to download and execute the skills package and updates from the NPM registry. This introduces a dependency on external, unversioned code executed during the skill's initialization phase.
[COMMAND_EXECUTION]: The skill relies heavily on the calibrate CLI tool to perform actions such as listing agents, resolving names, and creating test suites. This includes interpolating user-provided arguments and command outputs into subsequent shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external CLI command outputs (e.g., calibrate agents list, calibrate tests get). If the external service or tool output is compromised, it could inject malicious instructions into the agent's context. The skill lacks explicit sanitization for these ingestion points.
[OBFUSCATION]: The skill explicitly instructs the agent to perform the update and reload steps "silently," without narrating or mentioning them to the user, which is a form of behavioral concealment.

build-test-suite