The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation includes an installation command npx skills add Pwnote/skills which executes a shell command to install the skill from the vendor's repository.
[CREDENTIALS_UNSAFE]: The JSON schema defined in the skill includes a specific block type for credentials with fields for username and secret, which are intended to hold sensitive information.
[PROMPT_INJECTION]: The skill facilitates the processing of complex JSON files containing numerous free-text fields (such as markdown, poc, and description), creating a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent's context through user-supplied JSON files structured according to the pwnote engagement format (SKILL.md).
Boundary markers: There are no instructions or delimiters provided to help the agent distinguish between the structured data and potential embedded commands.
Capability inventory: The skill is designed to enable the agent to generate, edit, and verify these files, which involves reading and writing potentially malicious content.
Sanitization: The skill does not provide methods for sanitizing or escaping the content within the text-heavy fields.

pwnote-engagement-file