The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to generate a new verification skill (a SKILL.md file) containing shell commands and then immediately execute it end-to-end to prove it works.
Evidence: SKILL.md sections "Generate" and "Prove the generated skill" describe the workflow of writing and then running a new skill file.
[INDIRECT_PROMPT_INJECTION]: The skill populates the generated verification instructions using content discovered by interviewing the repository. If the repository contains malicious instructions disguised as launch or test commands, they could be incorporated into the execution pipeline.
Ingestion points: Repository-native commands, readiness signals, and environment configurations discovered during the "Interview the repository" phase in SKILL.md.
Capability inventory: The skill has the capability to write files and execute shell commands (Launch, Doctor, Drive, Cleanup) across the generated skill.
Boundary markers: No explicit delimiters are used to wrap the discovered repository content to prevent it from influencing the agent's behavior.
Sanitization: The skill relies on the agent's ability to identify "repository-native" tools but lacks specific filtering or validation of the strings used to build the commands.
[COMMAND_EXECUTION]: The skill involves the execution of various shell commands throughout the lifecycle of the generated verifier, including app launch, identity verification (Doctor), and system cleanup.
Evidence: Sections in SKILL.md and examples in references/feature-map-example/README.md describe running browser harnesses and terminal actions through control-notes cli.

create-verification-skill