Reservations Hybrid Advisor

Recommend optimal Azure Reservations and Hybrid Benefit coverage for maximum stacked savings

This skill appears coherent: it asks you to provide exported Azure reports or CLI output and explicitly avoids requesting credentials. Before using it: (1) do not paste any secrets, tokens, or full access keys — remove or redact any fields that look like keys or tokens; (2) sanitize PII or subscription IDs if you don’t want them shared; (3) prefer exporting read-only reports (CSV/JSON) from the Portal or using the az CLI locally with a least-privileged account (Cost Management Reader + Reader) and then paste the results; (4) confirm the tool’s claimed savings percentages and recommendations in the Azure Portal or with an authorized Azure partner before purchasing reservations; (5) ask the skill to explicitly confirm it will not attempt remote access if you are concerned about the 'bash' tool listed in the header. If you need higher assurance, run the analysis offline (locally) or share only aggregated/sanitized data.

Type: OpenClaw Skill Name: reservations-hybrid-advisor Version: 1.0.0 The `SKILL.md` file declares `tools: bash`, granting the agent shell execution capabilities. Although the skill explicitly states it is 'instruction-only' and 'does not execute any Azure CLI commands', it provides `az` CLI commands (`az consumption usage list`, `az vm list`) as examples for the user. This combination creates a potential prompt injection vulnerability where a user could instruct the agent to ignore its own safety rules and execute these read-only commands, or potentially other commands, despite the skill's stated benign intent. This represents a risky capability without clear evidence of intentional malice from the skill itself.