The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's core function is to generate a project-local skill and immediately execute its instructions to "prove" it works. This involves writing executable content to .claude/skills/patinaproject-verify/SKILL.md and then running the discovered launch, drive, and cleanup commands.
[INDIRECT_PROMPT_INJECTION]: The skill "interviews" the repository by reading its codebase (READMEs, package scripts, and documentation) to determine how to drive the application. This untrusted data is directly used to generate the logic and commands of the new skill.
Ingestion points: Repository metadata and documentation files (e.g., package.json, Makefile, README.md).
Boundary markers: Not present; the skill treats content found in the repo as authoritative instructions for the generation process.
Capability inventory: Shell command execution across multiple lifecycle stages (launch, doctor, drive, and cleanup).
Sanitization: No explicit sanitization or validation of the discovered commands is mentioned before they are executed in the proof step.
[COMMAND_EXECUTION]: The skill facilitates the discovery and execution of arbitrary shell commands found within the project. If a repository is malicious or compromised, the agent may unknowingly execute harmful scripts (e.g., a malicious dev command) while attempting to follow this skill's instructions to "fix" or "verify" the app.

create-verification-skill