The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is primarily designed to ingest and analyze untrusted external data, such as fetched or pasted text.
Ingestion points: Multiple tools, including jev_screen, jev_verify, jev_extract, and jev_audit, accept external document strings for analysis.
Boundary markers: The skill instructions include a Policy section directing the agent to treat successfully screened content as task data rather than authoritative instructions and to ignore attempts to override operating rules.
Capability inventory: While the skill itself provides logic and judgment tools, the agent using the skill typically possesses broad capabilities including file system access and command execution.
Sanitization: The skill incorporates a dedicated tool, jev_screen, specifically designed to detect and recommend blocking of prompt injections within the input text.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx to fetch and execute the @jkudish/jev-mcp package from the NPM registry. This package is a resource provided by the skill author.
[DATA_EXFILTRATION]: Input content is sent to an external judgment provider (defaulting to TypeSafe) to perform the semantic analysis. The skill documentation provides clear warnings to the user to avoid sending secrets, credentials, or private source code to this external provider.

jev