har-to-api-client
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes HAR (HTTP Archive) files which are containers for raw network traffic from external, potentially untrusted websites. This represents a significant attack surface where a malicious site could embed instructions within its JSON responses to influence the agent's behavior when the capture is analyzed.
- Ingestion points:
capture.haris ingested and processed in multiple steps (inventory, schema inference, code generation). - Boundary markers: The instructions do not specify any delimiters or ignore-instructions to protect the agent from data within the HAR file.
- Capability inventory: The skill executes shell commands via
harreconand writes files to the local system (client.py). - Sanitization: No explicit sanitization of the HAR content is mentioned for the agent's context, though the
harrecontool reportedly masks some auth samples. - [DATA_EXFILTRATION]: HAR files by definition contain sensitive session data such as cookies, Bearer tokens, and CSRF headers. Although the skill mentions masking and hygiene, there is a risk of accidental exposure if the agent outputs raw capture data to the chat history or logs, or if the generated client code inadvertently includes hardcoded secrets from the capture.
- [COMMAND_EXECUTION]: The skill relies on a specific CLI utility named
harreconfor its core functionality. The security of the skill's execution environment depends on the safety and provenance of this tool.
Audit Metadata