The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes output from the Prism CLI and local network services to generate user summaries, which creates a surface for indirect prompt injection if those sources return adversarial content.
Ingestion points: Data is ingested from the output of prism status --message (specified in the blueprint prompt) and from curl requests to http://127.0.0.1:18790/positions and http://127.0.0.1:18790/decisions.
Boundary markers: The skill's instructions for the hourly blueprint do not include specific delimiters or instructions to ignore embedded prompts when summarizing the trading status.
Capability inventory: The skill has the ability to execute terminal commands (prism, curl) via the required terminal toolset.
Sanitization: The instructions do not specify any validation or sanitization steps for the data retrieved from the local trading tools.
[COMMAND_EXECUTION]: The skill uses terminal-based commands to configure, start, and monitor the Prism trading agent.
Evidence: Commands such as prism setup, prism dev, and curl are used to interact with the local environment. These are expected behaviors for a skill that manages a local trading service using the terminal toolset.

prism-hermes