The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the parallel-pp-cli binary using npx -y @mvanhorn/printing-press-library and go install github.com/mvanhorn/printing-press-library/.... These resources originate from the identified vendor's official repositories and registry entries.
[COMMAND_EXECUTION]: The skill primarily functions by executing the parallel-pp-cli tool. It includes instructions for backgrounding commands (using &) and managing local state, which involves reading and writing to a local SQLite database.
[DATA_EXFILTRATION]: The CLI supports a --deliver webhook:<url> flag that allows command results to be POSTed to an arbitrary external URL. This represents a significant data exfiltration surface if an agent is influenced to route sensitive outputs, such as account balances or API keys generated via the service account-create-key command, to an attacker-controlled endpoint.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted data from the web via the extract, websearch, and findall commands.
Boundary markers: No explicit sanitization or boundary markers are mentioned for processing the markdown content returned from web extractions.
Capability inventory: The skill can write to files (--deliver file:<path>), make network requests (webhook:<url>), and execute system commands.
Sanitization: The skill includes a 'learning loop' (teach, recall, playbook amend) that stores information from previous sessions in a local SQLite database. Malicious content found during web research could be 'taught' to the system, potentially poisoning future research cycles or agent behaviors via persistent stored instructions.
[CREDENTIALS_UNSAFE]: While the skill correctly advises against committing API keys and mentions storing them in a credentials.toml file, it includes commands for creating new API keys (account-create-key), which exposes the generated credentials in the command output.

pp-parallel