The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install external binaries and packages from vendor-controlled repositories.
Evidence: npx -y @mvanhorn/printing-press-library install hackernews --cli-only and go install github.com/mvanhorn/printing-press-library/library/media-and-entertainment/hackernews/cmd/hackernews-pp-cli@latest.
[REMOTE_CODE_EXECUTION]: The installation process involves executing code downloaded from remote repositories (NPM and GitHub) associated with the vendor.
Evidence: The use of npx and go install to fetch and run setup scripts or binaries from the mvanhorn vendor space.
[DATA_EXFILTRATION]: The CLI includes a --deliver webhook:<url> flag that allows sending command output to an arbitrary external URL, providing a mechanism for exfiltrating processed data.
Evidence: Documentation for the --deliver flag describing the webhook:<url> sink which POSTs output to a provided endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from Hacker News, which consists of untrusted user-generated content that could contain malicious instructions.
Ingestion points: Data is retrieved from Hacker News via commands like stories, search, pulse, and hiring (SKILL.md).
Boundary markers: None identified in the provided instructions or examples.
Capability inventory: The skill can execute shell commands (hackernews-pp-cli), perform network operations (fetching HN data, posting to webhooks via the deliver flag), and write to a local SQLite store.
Sanitization: No explicit sanitization or filtering of external content is mentioned before processing or display.

pp-hackernews