The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes an environment (Nika workflows) that processes untrusted inputs and performs sensitive actions.
Ingestion points: Data enters workflows via inputs: declarations or the nika:fetch tool as documented in references/language.md.
Boundary markers: The language enforces mandatory permits: blocks and utilizes a static analyzer (nika check) to enforce a default-deny policy as documented in SKILL.md and references/language.md.
Capability inventory: The platform supports subprocess execution via the exec: verb, network requests via nika:fetch, and filesystem operations via nika:write as documented in references/native-and-artifacts.md.
Sanitization: The engine employs taint tracking (law: taint) and structured data manipulation via nika:jq to prevent injection vulnerabilities as documented in references/native-and-artifacts.md.
[COMMAND_EXECUTION]: The skill guides the agent in using the nika CLI for workflow management and instructs on the use of the exec: verb within workflows for subprocess execution. Security is maintained through explicit command whitelisting and static analysis checks.
[DATA_EXFILTRATION]: The skill documents the use of the nika:fetch built-in tool for network operations and secrets: for credential management. It provides guidance on using egress: sinks to govern where sensitive data can be sent, ensuring that data movement is declared and governed.

nika-authoring