The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data retrieved from the Raindrop.io API (such as bookmark titles, notes, and highlights) which can be injected into the agent's context during search and synchronization tasks.
Ingestion points: Data is ingested via raindrop sync, raindrop raindrops search, and raindrop highlights get-all (SKILL.md).
Boundary markers: The skill instructions do not specify the use of delimiters or specific warnings to ignore embedded instructions within bookmark metadata.
Capability inventory: The agent can execute shell commands via the raindrop-pp-cli binary, perform network operations to the Raindrop API and user-defined webhooks, and write to the local filesystem using the --deliver file:<path> flag.
Sanitization: There is no evidence of sanitization or filtering of external content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external binaries and packages from public registries and repositories.
Fetches the @mvanhorn/printing-press-library package via npx for CLI installation.
Downloads and installs the raindrop-pp-cli and raindrop-pp-mcp binaries directly from the mvanhorn/printing-press-library repository on GitHub using go install.
[DATA_EXFILTRATION]: The CLI includes a --deliver webhook:<url> feature that allows the agent to POST command outputs to an arbitrary remote URL. While this is a functional capability for routing data, it presents a risk if the agent is directed to exfiltrate sensitive information or bookmark data to an unauthorized endpoint.
[COMMAND_EXECUTION]: The skill is designed to interact extensively with the local operating system by executing the raindrop-pp-cli binary. It includes instructions for running background processes (e.g., using & for the teach command) and managing local state via SQLite databases and configuration files.

pp-raindrop