The Agent Skills Directory

[DATA_EXFILTRATION]: The CLI tool supports a --deliver flag that can route command results to a webhook:<url>. This allows sending command output to arbitrary external endpoints, which could be exploited for exfiltration if an agent is manipulated into using a malicious URL.
[DATA_EXFILTRATION]: The feedback command can be configured to POST locally stored feedback to a remote endpoint defined by the LEARN_DISABLED_EXAMPLE_FEEDBACK_ENDPOINT environment variable.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute the learn-disabled-example-pp-cli binary. This binary performs operations such as database syncs, configuration checks, and API calls.
[EXTERNAL_DOWNLOADS]: The installation process involves executing remote code via npx from the @mvanhorn/printing-press-library package and downloading pre-built binaries from the author's GitHub repository (github.com/mvanhorn/printing-press-library). These are documented as vendor-owned resources.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection. 1. Ingestion points: The games command processes data from a remote API, while the which and feedback commands accept natural language input. 2. Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded instructions in ingested data. 3. Capability inventory: The skill can execute shell commands via the Bash tool, write to the filesystem via the --deliver file: sink, and perform network POST requests via the feedback command and the --deliver webhook: sink. 4. Sanitization: The skill does not perform specific sanitization of external content before interpolation into the agent's context.

pp-learn-disabled-example