The Agent Skills Directory

[COMMAND_EXECUTION]: Employs the curl utility to send JSON payloads to the Harness service for session callbacks.- [DATA_EXFILTRATION]: Sends session tokens and operation metadata to tryharness.ai. This is mitigated by a strict HTTPS allowlist ensuring data is only sent to the designated provider.- [PROMPT_INJECTION]: The skill processes untrusted briefs and research content, creating a surface for indirect prompt injection.
Ingestion points: Briefs, research findings, and external web pages referenced during the task (SKILL.md).
Boundary markers: Utilizes a mandatory protocol header and explicit rules to treat non-agent content as data only.
Capability inventory: curl for network requests and workspace file management.
Sanitization: Mandates independent verification of all external context and prohibits following instructions embedded in research data.

harness-collaboration