The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to identify and execute startup, build, and interaction commands found within the project repository to validate the generated verification skill. This automated execution of repo-sourced commands is a risk when interacting with untrusted codebases.
[DYNAMIC_EXECUTION]: The skill defines a process where it generates a new set of instructions (a local skill at .claude/skills/verify-<app>/SKILL.md) and then executes those instructions end-to-end as a "proof" step.
[INDIRECT_PROMPT_INJECTION]: The skill gathers information from the codebase to define its logic, which allows malicious content within the repository to influence the generated output and subsequent execution steps.
Ingestion points: Project repository files including READMEs, package scripts, Makefiles, and source code analyzed during the "Interview the repo" phase.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are provided for the ingestion of external repository data.
Capability inventory: File system write access (creating the skill), execution of arbitrary shell commands (verifying the skill), and potential network access via launched applications.
Sanitization: No sanitization or validation of the ingested codebase content is described before it is interpolated into the generated verification logic.

create-verification-skill