autonomous-agent-harness
Pass
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill configuration guides the user to install official packages from Anthropic, including
@anthropic/memory-mcp-server,@anthropic/scheduled-tasks-mcp-server, and@anthropic/computer-use-mcp-servervia thenpxpackage runner. - [COMMAND_EXECUTION]: The documentation provides examples of shell commands for creating scheduled tasks and interacting with the programmatic interface of the agent (
claude -p). - [DATA_EXFILTRATION]: The skill references the official Anthropic dispatch API (
api.anthropic.com) for event-driven workflows. This is a well-known service and the usage is consistent with the skill's stated purpose of remote agent triggering. - [PROMPT_INJECTION]: The skill establishes an autonomous workflow that ingests untrusted data from external sources such as GitHub pull requests, notifications, and search results.
- Ingestion points: GitHub PR bodies, notifications, and Exa search results processed by the scheduled reviewer and research agent workflows.
- Boundary markers: None explicitly defined in the provided prompt templates.
- Capability inventory: The agent has capabilities to write to local memory files, interact with the file system, and use browser automation via MCP tools.
- Sanitization: No specific sanitization or filtering logic is provided in the example prompts to handle potentially malicious instructions embedded in the external data.
Audit Metadata