The Agent Skills Directory

[DATA_EXFILTRATION]: The CLI tool includes a --deliver webhook:<url> flag. This capability allows the agent to transmit sensitive data, such as Amazon order history, gift card balances, and transaction details, to external servers.
[CREDENTIALS_UNSAFE]: The skill manages Amazon session cookies to authenticate requests. It provides commands for exporting (auth export) and importing (auth import) these credentials. While the documentation suggests secure handling via secret managers, the existence of these commands and the use of the AMAZON_COOKIES environment variable create a surface for potential credential exposure.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external data from Amazon (e.g., item titles, seller names, tracking notes). This content is not controlled by the user or the skill author and could contain malicious instructions that trigger when the agent analyzes order history or search results.
Ingestion points: amazon-orders-pp-cli output (JSON results containing order details from Amazon).
Boundary markers: The skill uses a JSON provenance envelope to wrap command results.
Capability inventory: The agent has access to Read and Bash tools, allowing it to execute the CLI and process its output.
Sanitization: No explicit sanitization or filtering of order content is mentioned before the data is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install binaries using npx and go install from repositories associated with the vendor (mvanhorn). These are recognized as vendor-owned resources.

pp-amazon-orders