The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install the necessary tools using npx -y @mvanhorn/printing-press-library, which fetches code from the @mvanhorn organization on the npm registry.
[DATA_EXFILTRATION]: The CLI includes a --deliver webhook:<url> feature that enables sending the output of any command to an arbitrary network endpoint. Additionally, the feedback command can be configured to automatically POST data to a remote URL specified in OPENROUTER_FEEDBACK_ENDPOINT.
[DYNAMIC_EXECUTION]: The skill implements an 'Automatic learning' loop that journals all invocations and synthesizes 'playbooks' and 'candidates'. These are later replayed as shell commands with slot-based argument substitution, creating a mechanism for dynamic command generation based on past activity.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection by ingesting untrusted data from the OpenRouter API and user queries, which are then used to drive the command discovery (which command) and the learning/playbook synthesis system.
Ingestion points: User queries, OpenRouter API responses (model lists, activity logs, usage stats).
Boundary markers: No specific boundary markers or instruction-guarding delimiters are documented for data interpolated into command discovery.
Capability inventory: Execution of shell commands via the Bash tool, network communication (API and webhooks), and file system access (credentials and local journaling).
Sanitization: The documentation mentions a PII stripping process for taught queries, but notes that it relies on pattern matching and may not be comprehensive.
[COMMAND_EXECUTION]: The skill's primary function is to orchestrate the openrouter-pp-cli binary, involving numerous subcommands that manipulate the local file system (config, credentials, SQLite database) and interact with external APIs.

pp-openrouter