analyze-costs

Pass

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill follows least-privilege principles by utilizing a specific set of tools provided by the official Harness MCP server (harness-mcp-v2).
  • [SAFE]: Operations are restricted to read-only cost data retrieval (harness_get, harness_list) and limited feedback reporting for anomalies (harness_execute), preventing unauthorized system modifications or broad command execution.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from external cloud providers (resource names, service descriptions, cost metadata).
  • Ingestion points: External data enters the context via harness_get and harness_list calls in SKILL.md.
  • Boundary markers: The report format uses structured Markdown headers and tables to separate data sections.
  • Capability inventory: Limited to cost-specific MCP tool calls; no general shell access or network exfiltration capabilities are present.
  • Sanitization: None explicitly defined in the prompt, but the risk is negligible as the source data consists of system-generated billing records and optimization metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 18, 2026, 11:24 PM
Security Audit — agent-trust-hub — analyze-costs