haah
Pass
Audited by Gen Agent Trust Hub on Apr 8, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXFILTRATION]: The skill accesses an API key stored in
~/kyp/haah/haahconfig.ymlto authenticate requests toapi.haah.ing. It transmits natural language queries and receives responses from other users, which is the primary purpose of the communication service. - [PERSISTENCE_MECHANISMS]: The instructions prompt the agent to ask the user for permission to install a cron job (
*/30 7-22 * * *) or modifyHEARTBEAT.mdto ensure the skill runs regularly. This is a legitimate requirement for a messaging/dispatch system. - [COMMAND_EXECUTION]: Uses standard shell utilities including
curlfor network requests andyqfor parsing YAML configuration files. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: External natural language content is ingested via
GET /dispatchandGET /inboxfrom the vendor API. - Boundary markers: None provided in the instructions to delimit external content from system instructions.
- Capability inventory: The skill has access to the local file system (config) and shell execution (
curl,yq). - Sanitization: No specific sanitization or validation of the remote text is mentioned before the agent reasons over it.
Audit Metadata