The Agent Skills Directory

[PERSISTENCE]: The skill writes to ~/.cursor/rules/pstack-models.mdc and includes the alwaysApply: true frontmatter. This mechanism ensures that the model override instructions are loaded into the agent's context for every new session, creating a persistent behavioral modification.
[DYNAMIC_EXECUTION]: The skill generates executable instructions (in the form of a .mdc rule file) based on a template. This generated content is then dynamically interpreted by the agent platform to control which models are used for specific subtasks.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to injection through the model slug input process.
Ingestion points: The skill ingests data from ~/.cursor/rules/pstack-models.mdc and accepts manual text input from users to define model slugs.
Boundary markers: While the skill uses a specific template, it does not implement strict delimiters or 'ignore' instructions to prevent user-supplied slugs from containing malicious markdown or frontmatter overrides.
Capability inventory: The skill has the ability to write files to the agent's rule directory and trigger other agent capabilities like /create-verification-skill.
Sanitization: The instructions do not define sanitization or validation steps to ensure model slugs are limited to safe character sets (e.g., alphanumeric and hyphens), potentially allowing the injection of control characters to manipulate the resulting rule file structure.
[COMMAND_EXECUTION]: The skill performs file system operations, including reading and overwriting a configuration file in the user's home directory to manage the state of the pstack configuration.

setup-pstack