The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install executable code using npx and go install from repositories and packages controlled by the vendor (@mvanhorn/printing-press-library and github.com/mvanhorn/printing-press-library).
[DATA_EXFILTRATION]: The CLI supports an --deliver webhook:<url> flag, which allows the agent to transmit command output, including potentially sensitive character snapshot data, to an arbitrary external URL.
[INDIRECT_PROMPT_INJECTION]: The 'Automatic Learning' loop (recall, teach, playbook) enables the CLI to store and retrieve 'playbooks' and 'learnings' derived from previous interactions. This creates a stateful surface where untrusted data processed in one session (e.g., from a D&D Beyond page or character file) can influence the agent's future command choreography and decision-making.
Ingestion points: User-supplied queries, character snapshots (PDF/JSON), and public D&D Beyond rules pages processed by the CLI.
Boundary markers: The skill lacks explicit instructions for the agent to ignore or sanitize embedded instructions within processed data before 'teaching' the CLI.
Capability inventory: The skill executes a local binary (dndbeyond-pp-cli) which has access to the filesystem (character files) and the network (D&D Beyond API and webhooks).
Sanitization: There is a manual instruction (PII rule) to strip personal identifiers before teaching, but no automated sanitization is described for the learning loop.
[COMMAND_EXECUTION]: The skill is designed to drive a local CLI binary and involves shell command execution for installation, configuration, and operational tasks.

pp-dndbeyond