The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses prava CLI commands (setup, status, sessions create, sessions poll, mandate create, etc.) to manage payment workflows. It also uses find to locate the binary and npx for skill updates. These are restricted to the functionality of the Prava payment stack and follow a strict manual confirmation protocol for financial transactions.
[EXTERNAL_DOWNLOADS]: The skill references and directs the user to install the @prava-sdk/cli package from the official npm registry. It also provides an npx command to fetch other skills from the official Prava-Payments GitHub repository. Both sources are well-known or vendor-owned and are documented neutrally for the user to execute.
[INDIRECT_PROMPT_INJECTION]: The skill processes purchase data (merchant names, product descriptions, URLs) which could theoretically contain injection attempts. However, it mitigates this by requiring a 'HARD STOP' manual confirmation of the merchant and amount by the user before any transaction is initiated, and uses structured JSON for product data passed to the CLI.
Ingestion points: Purchase context (merchant name/URL, product descriptions) finalized in SKILL.md through discovery.
Boundary markers: The skill explicitly instructs the agent to present a human-readable confirmation to the user to prevent autopilot approval of malicious payloads.
Capability inventory: Execution of prava CLI commands via shell, npx skill additions/updates, and browser automation for checkout (as described in SKILL.md).
Sanitization: CLI commands use structured flags; however, product descriptions are passed as JSON strings which relies on the downstream CLI for final sanitization.

prava-pay